Nucl - Phys.B v.789

Nuclear Physics B 789 (2008) 144
Glueballs vs. gluinoballs: Fluctuation spectra in

non-AdS/non-CFT
Marcus Berg a, , Michael Haack b , Wolfgang Mck c
a Max Planck Institute for Gravitational Physics, Albert Einstein Institute, Mhlenberg 1, 14476 Potsdam, Germany
b Ludwig-Maximilians-Universitt, Department fr Physik, Theresienstrasse 37, 80333 Mnchen, Germany
c Dipartimento di Scienze Fisiche, Universit degli Studi di Napoli Federico II and INFN, Sezione di Napoli,
via Cintia, 80126 Napoli, Italy

Received 30 March 2007; received in revised form 17 July 2007; accepted 20 July 2007
Available online 27 July 2007
Abstract
Building on earlier results on holographic bulk dynamics in confining gauge theories, we compute the
spin-0 and spin-2 spectra of the MaldacenaNuez and KlebanovStrassler theories, that have non-singular
supergravity duals. We construct and apply a numerical recipe for computing mass spectra from certain
determinants. In the KlebanovStrassler case, the spectra of states containing the glueball and gluinoball are
universal and quadratic in the supergravity approximation, i.e., their mass-squareds depend on consecutive
number as m2 n2 for large n, with a universal proportionality constant. The hardwall approximation
appears to work poorly when compared to the unique spectra we find in the full theory with a smooth
cap-off in the infrared.
2007 Elsevier B.V. All rights reserved.
1. Introduction
One of the first successes of gauge/string duality was the computation of mass spectra of
strongly coupled gauge theories from dual geometries [13]. These original papers were all
concerned with N = 0 black hole solutions, where it is hard to check the validity of the correspondence. Other work uses the N = 4 superconformal theory (AdS bulk) plus a hard IR cutoff
[4] that serves to imitate interesting field theory IR effects, but there is no running coupling.
Studying N = 1 super-YangMills theory coupled to matter seems a good compromise in many
* Corresponding author.
E-mail address: mberg@aei.mpg.de (M. Berg).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.07.012
M. Berg et al. / Nuclear Physics B 789 (2008) 144
respects: these theories exhibit confinement, chiral symmetry breaking, running couplings and a
rich set of mass spectra.
For this non-AdS/non-CFT correspondence,1 where the bulk is not AdS and the boundary
field theory is not a CFT, much less is known about mass spectra, let alone correlators. A few
brave attempts exist in the literature [710], but as we explained in [11], their results on mass
spectra are at best inconclusive. In the present paper, we will compute the mass spectrum of the
N = 1 KlebanovStrassler theory [12], a non-conformal deformation of the N = 1 Klebanov
Witten theory [13] involving gluons and gluinos coupled to two sets of chiral superfields with a
specific quartic superpotential.2 We also compute the mass spectrum of the MaldacenaNuez
theory [16] (for which the supergravity solution was found in [17,18]), as a warmup. In this way,
we will be able to address physical questions about how confinement works in these models,
such as whether the theory displays linear confinement (for a recent discussion, see [19]).
Despite this promising state of affairs, many authors have emphasized that the aforementioned
theories are still quite far from real-world QCD. For instance, there are no open strings corresponding to dynamical quarks, hence no real meson spectra. On the good side, in [20,21], probe
D-branes were added to the KlebanovStrassler background, something which could be further
developed using our methods. Real QCD is also nonsupersymmetric, and we make heavy use of
the existence of a superpotential W . However, as indicated by the quotation marks, this superpotential is fake (in the sense of [22]), but there are nonsupersymmetric examples where such
structure exists irrespective of supersymmetry per se, see, e.g., [23]. Last but not least, real QCD
is not at large t Hooft coupling ; this is very difficult to
overcome with the present state of the
art, but the recent progress in [24] at least shows that 1/ corrections may not be unrealistic to
obtain for these theories.
Here, we take a step back from the effort to describe real QCD, and as mentioned above,
focus on an example that is well defined from a holographic point of view and see what can be
understood about that theory. We introduce some new techniques, but conceptually the biggest
difference from most of the literature is in things we do not do. We do not employ the hardwall approximation (which amounts to taking AdS as in Fig. 1 but with a finite IR cutoff). One
argument that has been put forward to motivate the use of the hard-wall AdS model is that some
physics should be insensitive to IR details. However, for computing mass spectra, a boundary
condition must be imposed in the IR, so this question cannot really be insensitive to IR details
(see, e.g., [24]). We discuss to what extent it is in Section 5.7. Another common approximation,
for example in the context of inflation [25], is the singular KlebanovTseytlin background [26],
as in Fig. 1. We insist on using the full KlebanovStrassler solution and imposing consistent
boundary conditions, and find that the commonly used approximations would likely not have led
to correct results even for this theory, let alone for QCD. It seems reasonable to ask for explicit
computational strategies to be developed concurrently with the search for real QCD, and we
believe we have made progress on such strategies here.
It is important to recall that the KlebanovStrassler theory does not have a Wilsonian UV
fixed point, but one can impose a cutoff in the UV to define the theory. One can think of this in
1 As far as we know, this expression was first used by Strassler at the Strings 2000 conference. It also featured promi-
nently in the titles of Aharonys 2002 [5] and Zaffaronis 2005 [6] lectures.
2 More precisely, we consider only mass eigenstates that are dual to bulk solutions in the KlebanovStrassler background, i.e., we do not consider meson mass spectra, which would arise from introducing flavor branes. Moreover, there
is always an issue with contamination by KaluzaKlein states in these theories, which we will ignore in the following.
For nice introductions to the KlebanovStrassler theory, see [14,15].
Fig. 1. Warp factors h(r). The string tension goes like h1/2 . From left to right the figures show the form of the warp
factor in AdS (tension goes to zero in infrared), KlebanovTseytlin (tension diverges at rs ) and KlebanovStrassler
(tension goes to finite value).
at least three ways: (1) this is an intermediate approximation and the theory will be embedded in
a more complicated theory with a UV fixed point like in [27]; (2) we will glue the noncompact
dual geometry onto a compact space that naturally provides a UV cutoff like in [28]; or (3) we
will want to match to (at least lattice) data at that point, and beyond that the theory is in any
case not asymptotically free (for recent ideas about this, see, e.g., [29]). As far as this paper is
concerned, any of these points of view may be adopted.
Now to the new results in this paper. We expand upon our earlier work [11] and present a
numerical strategy, the determinant method, to calculate the spectrum of regular and asymptotically subdominant bulk fluctuations. This puts us in the interesting position of being able to
compute the mass spectrum unambiguously, but not being able to say what the composition of
each mass state is without further information.3 This information should ideally come from data,
i.e., one should compute mixings at a specified energy scale where one has some control, and
use the dual theory to evolve into the deep nonperturbative regime. To illustrate this, let us take
the simpler example of a theory that does have a UV fixed point. The glueball operator Og , with
conformal dimension = 4, and the gluinoball operator Og , with = 3, have a diagonal mass
matrix in the conformal theory. As soon as we let the theory flow away from conformality, there
is a non-zero correlator Og Og . We can still diagonalize at any given energy, and the mixing
matrix will be energy dependent. For theories that are not conformal even in the UV, there is no
preferred basis labelled by the eigenvalues. One can still contemplate an approximate labelling
by (in the KS theory, this corresponds to expanding in the ratio P = (number of fractional D3branes)/(number of regular D3-branes)), but it is not yet completely clear how to implement this
in the dynamics; as we pointed out in [11], the limit P 0 does not commute with the UV limit
one needs for the asymptotics. We will give ideas about this in Sections 2.3 and 5.6, but we will
not resolve it completely.
The spectra we find have some simple features. The states of the KlebanovStrassler theory
come in towers, each of which shows a quadratic spectrum for large excitation number (i.e.,
m2 n2 , where n denotes the excitation number within the tower). This confirms the claims of
[19] also for the spin-0 states of the KS theory. However, for low excitation number (roughly for
the first two excitations in each tower) the structure of the mass values is richer.
For the MaldacenaNuez background, the spectrum we find is rather different. It has an
upper bound, in agreement with the analytical spectra that we reported in [11].
3 As will become clear, this is also true to some extent in standard AdS/CFT, but often ignored.
Let us also mention that there are other approaches to holography in this type of models,
such as the KaluzaKlein holography of [30]. Some issues may appear in a different light in that
framework, and that would be interesting to know.
The organization of the paper is as follows. In Section 2 we outline the general theory of
bulk fluctuations as well as the correspondence between their spectrum and the spectrum of mass
states in the dual gauge theory. The presentation is done such that it is applicable also in the
case of non-asymptotically AdS bulk spaces. Based on this general material, a numerical strategy to calculate the spectrum will be developed in Section 3. As a warmup, our first application
will be the MaldacenaNuez theory [16] in Section 4, before we come to the analysis of the
mass spectrum in the KlebanovStrassler cascading gauge theory in Section 5. Some conclusions can be found in Section 6, and many of the more technical details have been deferred to
Appendices AE.
2. Holographic mass spectra
We want to calculate mass spectra of confining gauge theories from the dynamics of fluctuations about their supergravity duals. In this section, we review and develop the main theoretical
tools that are necessary for such a calculation. We will start, in Section 2.1, with a review of the
treatment of the dynamics of supergravity fluctuations developed by us in [11], generalizing a
similar approach to the bulk dynamics in AdS/CFT [3133]. Then, in Section 2.2, we identify
the bulk duals of massive states in the gauge theory as eigenfunctions of a second-order differential operator satisfying suitable boundary conditions. Most of this material belongs to AdS/CFT
folklore, but we will fill in some details not commonly found in the literature. Finally, in Section 2.3, we will attempt to justify the identification of the gauge theory mass spectrum with the
spectrum of bulk eigenfunctions by studying the pole structure of holographic two-point functions. We hope that this goes some way towards a formulation of holographic renormalization
that holds beyond asymptotically AdS bulk spaces.
2.1. Linearized bulk dynamics
Let us start by reviewing the methodology we employ for studying fluctuations about
bulk backgrounds that are dual to confining gauge theories. We use consistent truncations of
type IIB supergravity that describe a subsector of the 10-dimensional dynamics in terms of a
5-dimensional effective theory. Details of this truncation can be found in our earlier paper [11].
Throughout this paper, we stick to a 5-dimensional effective bulk theory, but generalization to arbitrary dimension is straightforward. The effective theory is a non-linear sigma model of scalars
coupled to 5-dimensional gravity, with an action of fake supergravity type,

1
1
5
a b
S = d x g R + Gab () + V () .
(2.1)
4
2
The fake supergravity property4 implies that the scalar potential V () follows from a superpotential W () by
4 Here fake does not mean that the system is necessarily nonsupersymmetricthe systems we consider here are
supersymmetricjust that the formalism is applicable more generally. The relation between supergravity and fake supergravity was explored in [34,35].
1
4
V () = Gab Wa Wb W 2 ,
(2.2)
2
3
where we used the shorthand Wa = W/ a .
Background solutions, also known as Poincar-sliced domain walls or holographic renormalization group flows, are of the form
ds 2 = dr 2 + e2A(r) ij dx i dx j ,
a = a (r),
(2.3)
where the background scalars, (r),

and the warp function, A(r), are determined by the following
first-order (BPS-like) differential equations,
2
r A(r) = W (),
3
r a (r) = W a = Gab Wb .
(2.4)
Fluctuations about the Poincar-sliced domain wall backgrounds including the scalar and metric fields are best described gauge-invariantly. We will quote results from [11]. The physical fields
are vectors in field space and scalars in spacetime, denoted by aa , plus a traceless transverse tensor, eij , where i, j = 1, 2, 3, 4 are the indices along 4-dimensional Poincar slices. These fields
satisfy the following linearized field equations,

W a Wb 8
W b Wc
a
a
a
b
b
a 2A
b Dr + W |b
2 ac = 0,
W b c Dr W |c +
+ c e
W
3
W
(2.5)
and

8
r2 W r + e2A 2 eij = 0.
(2.6)
3
In (2.5) and (2.6), 2 = ij i j , which will be replaced by k 2 when we work in 4-dimensional
momentum space. Also, we use a sigma-model covariant notation, lowering and raising indices
of sigma-model tensors with Gab and its inverse Gab , and a sigma-model covariant derivative Da
that acts as
Da b a|b b a G c ab c ,
(2.7)
Gc
where ab is the Christoffel symbol derived from the metric Gab . Furthermore, we define a
background-covariant derivative Dr (r is the radial coordinate of (2.3)) acting as
Dr a = r a + G a bc W b c .
(2.8)
So, our task is to solve (2.5) and (2.6) for the backgrounds of interest, imposing appropriate
boundary conditions that we now describe.
2.2. Bulk duals of mass states
In AdS/CFT, the usual definition of the bulk dual of a boundary gauge theory mass state is a
solution of the (linearized) bulk field equations that is regular and integrable. In the following,
we will state these notions more precisely and present them in a form that is suitable also for
application in non-AdS/non-CFT.
As prototype for the equations of motion (2.5) and (2.6), we consider a system of n coupled
second-order differential equations of the generic form

8
2
2A(r) 2
Dr W Dr + M(r) e
(2.9)
k a = 0,
3
where Dr is the background-covariant derivative (2.7), M is a symmetric field-space tensor
(Mab = Mba ), and field indices have been omitted for the sake of brevity. We also assume that
the radial coordinate r is defined in a domain rIR r < , where r rIR corresponds to the
deep interior (IR) region, and large r to the asymptotic (UV) region of the bulk.5
Let us start with the regularity condition. As we want the 10-dimensional configuration obtained via uplifting of an effective 5-dimensional solution to be singularity-free, we must impose
that also the 5d solutions be regular in the bulk. For non-singular backgrounds, (2.9) typically
has a regular singular point at rIR and no others for finite r.6 Regularity in the bulk, therefore,
means regularity at rIR . To make sure that our regularity condition is invariant under changes of
field space variables, we require that the norm of the fluctuation vector is finite at r = rIR ,
a

Regularity condition:
a Gab ab r=r < .
(2.10)
IR
We emphasize that one cannot simply demand regularity of the components of aa at r = rIR ,
because r = rIR can coincide with a coordinate singularity in the sigma-model metric Gab , when
evaluated on the background. In the examples we consider, this indeed happens.
From the 2n independent solutions of (2.9), the condition (2.10) typically6 selects precisely n
independent regular solutions, which we will denote7 by areg,i , and n singular solutions, denoted
by asing,i . A generic solution of (2.9) is a linear combination of these,
a(r) = ci areg,i (r) + ci asing,i (r)
(2.11)
with constants ci and ci . Thus, (2.10) implies that

ci = 0 for all i.
(2.12)
Now, consider the integrability condition. Let us take a small detour and consider the bulk
Greens function G(r, r ; k 2 ), which satisfies

8
Dr2 W Dr + M(r) e2A(r) k 2 G r, r ; k 2 = e4A(r) (r r ),
(2.13)
3
where the factor e4A on the right-hand side is the metric factor 1/ g from the covariant delta
functional. The Greens function can be written in terms of a basis of eigenfunctions,

a (r)a (r )
G r, r , k 2 =
,
k 2 + m2
(2.14)
where the functions a satisfy (2.9) for k 2 = m2 . (Again, we omit the matrix indices, and the
indices of the two a s are not contracted.) Substituting (2.14) into (2.13) yields the completeness
5 We allow r = , like for example in a pure AdS bulk.
IR
6 By typically we mean that this is the case in all regular holographic configurations known to us. It would be
interesting to find more precise conditions for these statements.

7 We hope there will be no confusion between the index i = 1, . . . , n and the field space index a.
relation

a (r)a (r ) = e2A(r) (r r ),
(2.15)
from which one can deduce the orthogonality relation

dr e2A(r) a (r) a (r) = .
(2.16)
With the dot product we denote the covariant contraction of indices. Eq. (2.16) provides the
condition for the eigenstates a to be integrable. Due to the factor e2A (for general dimension d
it would be e(d2)A ), the integral measure is not the covariant bulk integral measure that one
might have expected.
Now, the spectrum of bulk eigenfunctions a is identified with the mass spectrum of operators dual to the fields a. (We will also say that the eigenfunction a is the bulk dual of a state of
mass m .) Clearly, this spectrum depends on the boundary conditions one imposes. One boundary condition is given by the regularity condition (2.10), the other follows from (2.16), namely

Integrability condition:
dr e2A(r) a(r) a(r) < .
(2.17)
For large r, i.e., in the UV, (2.9) has 2n independent asymptotic solutions. Let us denote the n
dominant ones by adom,i , and the n subdominant ones by asub,i (by dominant we mean leading
in r). Again, a generic solution of (2.9) can be written as a linear combination of these,
a(r) = di asub,i (r) + di adom,i (r),
(2.18)
with constants di and di .
It can be checked in the various cases we consider that the asymptotically dominant behaviors, adom,i , are not integrable in the sense of (2.17), whereas the subdominant behaviors are
integrable. Thus, (2.17) is equivalent to
di = 0 for all i.
(2.19)
To summarize, the eigenfunctions a , interpreted as the bulk duals of boundary field theory
states with mass m , are such that
ci = di = 0,
i = 1, 2, . . . , n,
(2.20)
from which follows that

a (r) = c,i areg,i (r) = d,i asub,i (r).
(2.21)
2.3. Pole structure of holographic 2-point functions

The masses of states (particles) in a quantum field theory manifest themselves as poles in the
two-point functions of operators, if there is a non-zero probability that these states are created by
the operators from the vacuum,8
0|O1 (0)||O2 (0)|0

d4 x eikx O1 (x)O2 (0) =
+ c.t.
(2.22)
k 2 + m2
8 Typically, terms that are formally infinite and analytic in k 2 (contact terms, c.t.) are needed to make the sum over
the spectrum convergent. The finite parts of these counterterms are renormalization scheme dependent. For a simple
example, see Appendix A.
For comparison with this general formula, we will now try to obtain the pole structure of holographic 2-point functions from the linearized dynamics of the bulk fields that we described in
Section 2.1. But we must start with a disclaimer. As we have not systematically dealt with the
renormalization and dictionary problems (see the discussion in [11]), our results will depend on
two strong assumptions, motivated by how things work in AdS/CFT. A better understanding of
these assumptions as well as the meaning of cases in which they are not satisfied would be very
desirable.
Let us start with the asymptotic expansion of the bulk fields (2.18). For a given bulk field a,
the coefficients di and di clearly depend on the choice of basis functions asub,i and adom,i , respectively. In order to remove some of the ambiguities, we assume that the term containing k 2
in the equation of motion (2.9) is asymptotically suppressed, so that the leading terms of the
asymptotic solutions are independent of k 2 , and the subleading terms can depend only on powers
of k 2 (assumption 1).9 Then, we can choose a basis of asymptotic solutions such that each has
a distinct leading behavior. These solutions are classified according to their asymptotic growth
with the radial coordinate into dominant and subdominant solutions. With each dominant asymptotic solution adom,i , we associate an operator Oi of the dual field theory.10 Two ambiguities still
remain in the choice of basis functions, but we now argue that these correspond to field theory
ambiguities. First, the normalization of the leading terms in the asymptotic basis solutions can be
freely chosen. In field theory this corresponds to a choice of normalization of the operators Oi ,
which always drops out in physical scattering amplitudes. Second, we have the freedom to add
to a given dominant solution multiples of asymptotic solutions of equal or weaker strength, with
coefficients polynomial in k 2 . (Solutions of equal strength can be added only with coefficients
independent of k 2 .) This is also expected from the dual field theory, since operators of a given
dimension can mix with operators of equal or lower dimensions (higher dimension operators being suppressed by our large UV cutoff) which implies that generically, there is no unique way
to define the renormalized operators. Therefore, the remaining ambiguities in the choice of an
asymptotic basis of solutions reflect the usual freedom in the definition of field theory operators.
For what follows, we do not need to make particular choices for the subdominant basis solutions, asub,i . This will be clearer once we obtain our final result of this section, i.e., (2.34).
Now for the second assumption. In a field theory calculation of correlation functions of the
operators Oi , one adds a source term to the Lagrangian,11
L = di Oi .
(2.23)
In holography, the sources are identified with the coefficients di of (2.18), whose dependence on
k 2 was suppressed in that equation. The second strong assumption we make is the generic form
of the exact one-point function of the operators Oi . In AdS/CFT, the response functions di of
(2.18) encode the exact one-point functions, up to scheme dependent terms that are local in the
sources [3639].12 In view of the ambiguities discussed above, we assume

Oi (k) exact = Yij k 2 dj k 2 + local terms,

(2.24)
9 This assumption is quite strong, and as we shall see, it holds in the KS system, but not in the MN system.
10 This does not make reference to components of the field space vector a(r), which would require the operator O to
carry a field index a and make it transform under bulk field redefinitions.
11 The minus sign is a useful convention. Let O
i exact be the exact 1-point function (i.e., the 1-point function in the
presence of finite sources dj ). Then, a connected (n + 1)-point function involving Oi is given by the nth derivative of
Oi exact with respect to the sources dj . With a plus sign, one would have an additional factor (1)n .
12 Imposing the regularity condition (2.10), only one set of coefficients is independent, which is taken to be d .
i
with a matrix Yij (k 2 ), which we will be able to determine further below (assumption 2). Part of
the assumption is that the poles of the connected 2-point functions arise in dj , and that Yij does
not give rise to additional poles. In AdS/CFT, (2.24) follows from holographic renormalization,
but unfortunately we have no proof of it in this more general setting.
We will now derive the general pole structure of holographic 2-point functions. To start, consider the general formula for a solution a(r, k 2 ) of (2.9) in terms of the Greens function and
prescribed boundary values. Let r0 be a (large) cutoff parameter determining the hypersurface
where the boundary values are formally prescribed. Remembering that neither the Greens function nor its derivative vanish at the cutoff boundary, we have13

a r, k 2 = e4A(r0 ) Dr0 a r0 , k 2 G r0 , r; k 2 a r0 , k 2 Dr0 G r0 , r; k 2 ,
(2.25)
where a(r0 , k 2 ) and Dr0 a(r0 , k 2 ) are the prescribed values of the field and its first derivative at
the cutoff boundary, respectively. Since r0 is an unphysical cutoff parameter, we must ensure that
the bulk field a(r) remains unchanged when r0 is varied. This is easily achieved, if, together with
a change of the cutoff, r0 r0 + r0 , the boundary values are changed by

Dr0 a r0 , k 2 = Dr20 a r0 , k 2 r0 ,
a r0 , k 2 = Dr0 a r0 , k 2 r0 ,
(2.26)
and the second derivative, Dr20 a(r0 , k 2 ), is determined by the equation of motion (2.9). To assure
(2.26), we determine the formal boundary values at the cutoff, a(r0 , k 2 ) and Dr0 a(r0 , k 2 ), from
the generic asymptotic behavior (2.18), with coefficients di and di fixed. After inserting (2.18)
and (2.14) into (2.25), we obtain
a (r)

a r, k 2 = e4A(r0 )
dl Dr0 asub,l r0 , k 2 a (r0 ) asub,l r0 , k 2 Dr0 a (r0 )
2
2
k + m

(2.27)
+ dl Dr0 adom,l r0 , k 2 a (r0 ) adom,l r0 , k 2 Dr0 a (r0 ) .
To continue, we observe that for very large r0 , the term on the second line of (2.27), containing
only subdominant solutions, is much smaller than the term on the third line. Therefore, we drop
it. Moreover, as we are interested only in the pole structure, we consider k 2 very close to m2
and expand the numerator keeping only the leading term, i.e., we replace k 2 by m2 in the
numerator. Finally, we use the fact that the eigenfunctions are purely subdominant,14

a (r) = d,i asub,i r, m2 .
(2.28)
This yields

a r, k 2

d,j d,i

asub,i r, m2 ,
dl Zlj m2
2
2
k + m
k 2 m2
=
(2.29)
13 This formula follows from (2.13) upon multiplication by e4A(r) a(r) from the left, taking the integral over r, integrating by parts and using the field equation (2.9). The IR boundary does not contribute, because e4A vanishes there.
The reason for this is that r = rIR should correspond to a single point of the bulk space, which is only guaranteed if e4A
vanishes there, cf. (2.3).
14 At this point one may wonder where the dominant part of a comes from. It arises from the sum over the spectrum in
(2.27), in particular from the UV contribution. For the simple case of AdS bulk, we show this in Appendix A. However,
it does not contribute to the poles.
10
with

Zij k 2 = e4A(r) Dr adom,i r, k 2 asub,j r, k 2 adom,i r, k 2 Dr asub,j r, k 2 .
(2.30)
Notice that, by virtue of the equation of motion (2.9), Zij is independent of r.
Thus, after reading off the response function di from (2.29), we obtain the poles of the connected 2-point function, using (2.24) and differentiating with respect to the source dj ,

d,i d,j

Oi (k)Oj (k) =
Yii m2 Zjj m2
+ c.t.
k 2 + m2
(2.31)
By symmetry, we find that Yij equals Zij up to normalization,

Yij m2 = N Zij m2 .
(2.32)
Therefore, defining also

Z,i = Zii m2 d,i

= e4A(r) Dr adom,i r, m2 a (r) adom,i r, m2 Dr a (r) ,
(2.33)
our final result is
Z,i Z,j
Oi (k)Oj (k) =
N
+ c.t.
k 2 + m2
(2.34)
We note that the Z,i are independent of the choice of the subdominant basis solutions, because
the normalization of the eigenfunctions a is fixed by (2.16).
The determination of the normalization factors N is an open problem, whose solution is
intimately related with finding a proof of (2.24). We conjecture that N = 1, as can be easily
checked for asymptotic AdS bulk spaces (see Appendix A).
3. Finding mass states
In the previous section, we characterized the duals of mass states as regular (in the IR) and
asymptotically subdominant (in the UV) solutions of the linearized bulk field equations. To calculate the spectrum, one solves a system of second-order differential equations with boundary
conditions specified at two endpoints. The standard numerical strategy for this is shooting, in
which one solves the corresponding initial value problem with two boundary conditions (field
value and derivative) specified at the initial point, and then varies the initial condition trying to
find the desired boundary value at the final point. For coupled multifield systems, this is a laborious process. Fortunately, since we are interested only in the spectrum and not in the explicit dual
bulk solutions, we can adopt a much simpler strategy. It involves computing a single function, the
determinant of a matrix, that depends on the momentum k 2 and the vanishing of which signals
the presence of a mass state.
In Section 3.1, we shall present the general idea of our strategy in its simplest implementation
involving the numerical solution of the field equations as an initial value problem starting close
to rIR . This simple implementation presents some numerical challenges, which we describe,
and which are overcome with the refined method described in Section 3.2. Further details of
numerical issues will be deferred to Appendix E.
11
3.1. Mass spectrum: Determinant method

We begin by imposing regularity in the IR. This means that we consider the equations of
motion for r rIR , find independent solutions of these, and impose ci = 0 in the language of
Section 2.2. Starting with these initial conditions, one numerically evolves to larger r to obtain the
n regular bulk solutions areg,i (r). For large r, the asymptotic behavior of each of these solution
can be expanded in terms of the UV-dominant solutions adom,j , so that
areg,i (r) j i adom,j (r)
(large r),
(3.1)
for some k 2 -dependent but r-independent constants j i . (The order of indices has been chosen
for later convenience. As a reminder, both areg,i (r) and adom,j (r) are n-component vectors in
field space, but we are suppressing the field-space index.) Thus, a general regular solution (cf.
(2.11)) will behave as
areg (r) ci j i adom,j (r)
(large r).
(3.2)
In order for areg to qualify as the bulk dual of a boundary field theory mass eigenstate, the
remaining condition in (2.20) is di = 0 for the fluctuation to be integrable. Thus,
di = ij cj = 0 for all i
mass state.
(3.3)
Regarding this as a set of n equations for the coefficients ci , it is necessary and sufficient for the
existence of non-zero solutions that the determinant of the matrix ij be zero:

det ij k 2 = 0 mass state with m2 = k 2 ,
(3.4)
where we have restored the dependence of the matrix ij on the 4-momentum k 2 .
To compute ij , let us make the field-space component index explicit in Eq. (3.1):
a
a

areg (z) i j i adom (z) j ,
(3.5)
where areg (z) and adom (z) are viewed as n n matrices, with column index i and row index a.
Inverting this equation, the coefficient matrix ij can be found by calculating15

ij (adom )1 (areg ) ij
(3.6)
for some large value of r, where the subdominant parts of areg are sufficiently suppressed. We
would like to emphasize that formula (3.6) renders the coefficients ij independent of the coordinates in field spacethe component index a is contracted. Hence, also the mass condition (3.4)
is independent of the choice of sigma-model variables.
However, as announced above, the IR UV determinant method does not work straightforwardly for coupled multifield systems in practice. The method is only sufficiently insensitive to
numerical error if all dominant solutions adom,i are comparable in magnitude, such that the integration error from the stronger ones does not swamp the significant digits of the weaker ones. For
1-component systems, this is obviously never a problem, and it turns out also not to be a problem
for the 2-component system in the MN background (Section 4.3). In general, and in particular
for the 7-component spin-0 system in the KS background, it is better to tread more carefully and
only evolve to a prescribed midpoint value, as we will describe in the next subsection.
15 This equation is intended in the sense of matrix multiplication, so there is no G involved.
ab
12
One could have imagined that this numerical difficulty would be alleviated by evolving the
other way (UV IR). This is actually partially true, but the method in the next subsection is still
more useful (see Appendix E for a few more details). Indeed, imposing the relevant initial conditions at a large value of r, one can calculate numerically the n UV-subdominant solutions asub,i .
Decomposing their behavior at r rIR using (2.11) into
asub,i (r) = j i asing,j (r) + j i areg,j (r),
(3.7)
the same argument as before affirms that an asub dual to a mass eigenstate exists, if det ij = 0.
The coefficients ij can be determined by matching the components of asub to the generic singular behavior.
The build-up and growth of integration error on the way from the asymptotic region to rIR
is an issue here and must be resolved. This can be done by making sure the UV cutoff is not
too large. Then, the integration error, which generally grows like the asymptotically weakest
solution, remains small.
3.2. Mass spectrum: Midpoint determinant method
The midpoint method is numerically the most stable and combines the two previous approaches. One calculates the asymptotic solutions analytically both in the UV and in the IR.
Then, the n regular solutions areg,i are evolved from r rIR up to a mid-point rmid , and the n
subdominant solutions asub,i are evolved from large r down to rmid . Then, one tries to find linear
combinations of the regular solutions and of the subdominant solutions, respectively, such that
their value and first derivative at rmid match:

areg
c
asub
(3.8)
= 0.
r areg r asub r=rmid d
As before, the necessary and sufficient condition for the existence of a mass eigenstate is given by
the determinant of a matrix (this time 2n 2n) being zero. Put differently, rather than demanding
that a given individual field match smoothly across the midpoint, we allow for the situation that
only some linear combination of the basis fields matches smoothly, and this linear combination
is determined by the coefficients c and d.
If the solutions we are matching were approximate analytical solutions and the midpoint were
the classical turning point, this midpoint matching would be precisely the WKB approximation.
Of course, our solutions will not be approximate analytical, but numerical and exact (up to integration error), so it is no approximation in that sense.
4. MaldacenaNuez system
4.1. General relations
The effective 5d model describing the bulk dynamics of the MN system contains three scalar
fields (g, a, p) and is characterized by the sigma-model metric [11,40]
Gab a b = g g + e2g a a + 24 p p,
(4.1)
13
and the superpotential16

2

1/2
1
W = e4p a 2 1 e4g + 2 a 2 + 1 e2g + 1
.
(4.2)
2
Let us summarize what Poincar-sliced domain wall backgrounds this system admits. In what
follows, we shall denote the background values of the scalar fields by g, a and p, and for their
fluctuations we will use the gauge-invariant sigma-model vector a = (g, a, p)T .17 The background equations (2.4) are most easily solved in terms of a new radial coordinate, , defined
by
= 2e4p r .
(4.3)
We shall need only the explicit solutions for a and g, which are given by
a=
2
,
sinh(2 + c)

e2g = 4 coth(2 + c) a 2 + 1 .
(4.4)
The integration constant c, which appeared for the first time in [41], can take values 0 c .
The regular background solution corresponds to c = 0, while c > 0 leads to singular bulk geome = 0, from which one
tries. The solution (4.4) is defined for ,
with defined by e2g()
obtains the relation
coth(2 + c) =
1
1.
2
(4.5)
Hence, takes values between 0 and 1/4, with = 0 for c = 0 and = 1/4 for c = . For
c > 0, = is the location of the singularity.
For the scalar p we will only need its relation to the warp function A (see (2.3)), which is
e2A e8p = C 2 ,
(4.6)
where C 2 is an integration constant setting the 4d scale. For later convenience, we set C 2 = 1/4.
Let us turn to the fluctuation equations (2.5) and (2.6) in terms of the radial variable . After
using the relations for the background, one straightforwardly finds that (2.5) becomes

( M)( N ) k 2 a = 0,
(4.7)
where we dropped field indices, and the matrices M and N are given by

W a Wb
a
4p
a
,
b W
N b = 2e
W

M a b = N a b 4e4p G a bc W c W ba .
(4.8)
Note that M and N are independent of p. As was observed in [11], the fluctuations of p, which
are described by the component a3 , decouple from the other two. This can be seen from (4.8) as
follows. From (4.1) and (4.2) we find the identities
W3 = 4W,
1
W 3 = W,
6
p W a = 4W a ,
16 We have adjusted the overall factor of the superpotential of [40] to our conventions.
17 The superscript T denotes the transpose, because a is a column vector.
(4.9)
14
from which follow N a 3 = N 3 b = 0. Moreover, one easily checks that G 3 bc = G a 3c = 0, so that

M a 3 = 4e4p W 3a , and M 3 b = 4e4p W b3 . Thus, a3 satisfies

4e4p W k 2 a3 = 0.
(4.10)
This is just the equation for a massless scalar field in the domain wall background.
Similarly, (2.6) gives rise to

4e4p W k 2 eij = 0,
(4.11)
which is identical to (4.10).

4.2. Analytic solutions in singular background
We will briefly introduce the analytical solutions for the fluctuations in the singular background case c = derived in [11]. These analytical solutions give useful indications for glueball
and gluinoball masses that can be compared with the numerical analysis in Section 4.3. Further
details can be found in our earlier paper [11].
Let us start with (4.7). In the case c = , the matrices M and N are very simple and diagonal,

1
1
1
M=
diag 8 + 4 , 2 + , 8 ,
4 1

1 1
N = diag , 2, 0 .
(4.12)

With these expressions, (4.7) can be solved analytically. We shall illustrate the analysis considering the independent solutions for the second component,

(z)3/2 ( 54 32 , 52 ; z),
2
(+1/2)z
a e
(4.13)
( 14 32 , 12 ; z),
where is the confluent hypergeometric function of the first kind (or Kummer function M), and
and z are defined by18

1
1
1
1
= 1 + k2 ,
z=
(4.14)
.
2
4
The two solutions in (4.13) have different behavior at the singularity, i.e., for = 1/4.
Notice that both solutions satisfy the regularity condition (2.10), so we do not have the standard
means of excluding one of them. In other words, in the background with c = , the regularity
condition (2.10) does not uniquely specify the IR boundary value that should be imposed on
the eigenfunctions.19 Let us tentatively discard the solution with the leading small-z behavior of
(4.13), which will be justified by the numerical results in the next section. This leaves us with
the first solution. The generic asymptotically dominant UV behavior is absent in this solution, if
18 The combination 1/4 is simply ;
see the discussion around (4.5).
19 One can show that the same behavior is found for all c > 0 by expanding the matrices M and N for .
The
regular case c = 0 is qualitatively different from these, because the limits c 0 and do not always commute. To
see this, consider, e.g., da()/ d at = .
15
takes one of the values [11]

1
= (4n + 1),
6
n = 1, 2, 3, . . . .
(4.15)
As the second component is dual to the gluino bilinear, these states may be interpreted as
gluinoballs.
Applying the same argument to the solutions of the first component [11], we find states for
1
= (4n + 3),
6
n = 1, 2, 3, . . . .
(4.16)
The spectra (4.15) and (4.16) can be combined into

1
= (2n + 3),
6
n = 1, 2, 3, . . . ,
(4.17)
with odd and even n corresponding to (4.15) and (4.16), respectively. Using (4.14), we can
rewrite (4.17) as
m2 =
4n(n + 3)
9
=1
,
(2n + 3)2
(2n + 3)2
n = 1, 2, 3, . . . .
(4.18)
We would like to remark that, in [11], we had obtained a spectrum identical to (4.18), with the
addition of a massless gluinoball for n = 0. The above heuristic argument, that the solution with
leading small-z behavior should be discarded, disposes of the massless gluinoball. This choice
will be justified a posteriori by the numerical results.
As we move from the background with c = to the regular background with c = 0, we
expect these values to change. Moreover, in the regular background, no heuristic argument is
needed, as the regularity condition will exclude half the solutions.
From the solutions of the third component, no discrete mass values are found [11].
4.3. Numerical solutions in the regular background
Let us now consider (4.7) in the regular background, i.e., for c = 0. The matrices M and N are
considerably more complicated than in the previous case, and we list their entries in Appendix B.
Thus, the full equations of motion will be solved numerically.
For our purposes, the simple approach of Section 3.1 is sufficient for the MN system. (It will
not be sufficient for the KS solution.) Hence, let us first consider in the vicinity of = 0, and
find the regular solutions of (4.7). Expanding the matrices M and N about = 0 yields
2 +
4
M =
3
4
3
34
1
27
8
1
3
0
20
9
N = 4
0
0
1
3
1
+
+
0
14
15
8
9
0
0
3
+O ,
2 8
9
0
3
0 + O .
(4.19)
(4.20)
Recall from the discussion at the end of Section 4.1 that the zeros in (4.19) and (4.20) are exact.
16
Table 1
Comparison between numerical results (m2 ) and analytical estimates for the mass spectrum in the MN background
n
m2
(4.18)
n
m2
(4.18)
n
m2
(4.18)
1
0.4068
0.6400
9
0.9782
0.9796
17
0.9932
0.9934
10
0.9828
0.9830
18
0.9940
0.9941
2
0.8078
0.8163
3
0.8675
0.8889
11
0.9848
0.9856
19
0.9945
0.9946
12
0.9875
0.9877
20
0.9951
0.9951
4
0.9235
0.9256
5
0.9406
0.9467
13
0.9888
0.9893
21
0.9954
0.9956
14
0.9905
0.9906
22
0.9959
0.9959
6
0.9592
0.9600
7
0.9662
0.9689
15
0.9914
0.9917
23
0.9961
0.9962
16
0.9926
0.9927
24
0.9965
0.9965
8
0.9747
0.9751
Using (4.19) and (4.20), it is straightforward to solve (4.7) in terms of series expansions about
= 0. The independent regular behaviors are

1

1 2 8 2
k
a1 = 1 +
(4.21)
0 + O 4 ,
6
3
0

1

2
a2 = 3 + O 4 ,
(4.22)
0

0

1 2 2
a3 = 1 + k
(4.23)
0 + O 4 .
6
1
The other three solutions go as 1/ in the leading term and will be discarded.
The asymptotic region ( 1) is well described by the singular background considered in
Section 4.2. Hence, a convenient basis of generic dominant asymptotic behaviors can be found
from the analytical solutions of [11]. Combining them to a matrix, where each column is an
independent solution, we get
1/43/2
0
0
1
.
adom = e(1 2 )
(4.24)
0
0
1/43/2
0
1/4+/2
Putting into practice the method of Section 3.1, we have searched for zeros of det , where the
matrix is defined by (3.6), with adom given by (4.24), and areg being the matrix of independent
numerical solutions at large . In order to calculate them, it is convenient to rewrite (4.7) as a
system of first-order equations,20

a
a
N 1
.
(4.25)
=
b
b
k2 M
In order to calculate the three independent regular solutions, we impose the three initial conditions (4.21)(4.23) at very small .
Our results are as follows. For the system involving the first two components of a, we obtain
discrete mass values. The first few of these are listed in Table 1 and can be compared to the
20 Notice that both a and b contain 3 components, so that (4.25) is a system of 6 first-order ODEs. It is convenient
to consider the third component of a separately, as it fortuitously decouples from the other two. Thus, we have a 4component and a 2-component system of first-order ODEs.
17
Fig. 2. The function (4.26) for the numerical mass spectrum in the MN background.
estimates from the singular background, (4.18). To refine this comparison, in Fig. 2 we plot the
function
1/2 2

n,
f (m, n) = 1 m2
3
(4.26)
which leads to the mass values

m2 = 1
9
.
(2n + 3f )2
(4.27)
If (4.18) were exactly true, f (m, n) would be identical to 1. We find that the numerical values
consistently come out smaller than those from the analytical approximation, and that they cluster
into two distinct spectra. For even n, f 0.95, whereas for odd n, except for the first few values,
f 0.78. This appearance of two spectra is not too surprising, if we remember that, in (4.18),
values for even and odd n originated from the first and second component of a, respectively.
These two components couple in the regular background so that their spectra will combine into
one. However, the numerical results indicate that there are two different kinds of bound states in
this spectrum.
From the massless scalar equation (4.10), which describes both the third component of a and
the spin-2 field eij , we have not found any discrete mass states, confirming the analytical result
of Section 4.2.
5. KlebanovStrassler system
5.1. KS background
The effective 5d model describing the bulk dynamics of the KS system contains seven scalar
fields. We will use the PapadopoulosTseytlin [40] variables (x, p, y, , b, h1 , h2 ). In the following, we shall briefly summarize the general relations for this system and the KS background
solution. For more details, see our earlier paper [11] and references therein.
18
The sigma-model metric is

Gab a b
1
1
P 2 2x
= x x + 6 p p + y y + +
b b
e
2
4
2

1
+ e2x e2y (h1 h2 ) (h1 h2 ) + e2y (h1 + h2 ) (h1 + h2 ) ,
4
(5.1)
and the superpotential reads

W =
1
1 2p2x
+ e4p cosh y + e4p2x (Q + 2P bh2 + 2P h1 ).
e
2
4
(5.2)
Here, Q and P are constants related to the number of D3-branes and wrapped D5-branes, respectively.
It is useful to introduce the KS radial variable by
= e4p r .
(5.3)
Here and henceforth, a field variable denotes the background of that field (p in this case), while
the sigma-model covariant a will be used for fluctuations. In terms of , the KS background
solution of (2.4) is given by
= 0 ,
(5.4)
e = tanh(/2),
,
b=
sinh
Q
h1 =
+ P e0 coth ( coth 1),
2P
coth 1
,
h2 = P e0
sinh
2 6p+2x
= coth
e
,
3
sinh2
(5.5)
(5.6)
(5.7)
(5.8)
(5.9)
e2x/34p = 2P 2 e0 32/3 h( ) sinh4/3 ,
(5.10)
with

h( ) =
coth 1
2
sinh
2 sinh(2) 4
1/3
(5.11)
Moreover, one can show that the warp function A satisfies

2/3
h( ),
e2A8p e6p2x sinh
(5.12)
where the proportionality factor depends on an integration constant that sets the 4d momentum
scale.
19
5.2. KS spin-2 spectrum

Let us start the numerical analysis of fluctuations in the KS background with the simplest
case: the free massless scalar describing the transverse traceless components of fluctuations of
the bulk metric. Specifically, we are interested in the spectrum of spin-2 glueballs, expanding on
earlier work by Krasnitz [7,42] (see also [43]).
The equation of motion (2.6), after substituting the appropriate relations from the previous
subsection, takes the form
2

+ 2e6p2x k 2 e2A8p e = 0,
(5.13)
where we omitted tensor indices on e. Defining
I ( ) =
h( )
h(0)
(5.14)
and appropriately choosing the integration constant in the warp function (5.12) such that

2/3
e2A8p = e6p2x sinh
(5.15)
I ( ),
we can rewrite (5.13) as

2/3
2
e = 0.
+ 2e6p2x k 2 I ( ) e6p2x sinh
(5.16)
The choice of momentum scale normalization in (5.15) is a little unusual, but convenient in what
follows.
In this simple one-component system, the method described in Section 3.1 is sufficiently
robust.21 For this method, first we need to find the regular behavior at = 0, from which the
initial conditions will be inferred. For small , we find to leading order e6p2x 1/ and
I ( ) 1, so that a regular solution of (5.16) is found to behave as

k2 2
(5.17)
+ O 4 ,
6
whereas the singular behavior, which starts with 1/ , is discarded. Notice that the choice of
momentum scale in (5.15) makes the k 2 term in (5.17) independent of h(0), a good thing since
the value of h(0) is known only from numerical evaluation of (5.11).
The second step of this method is to consider the asymptotic UV region. For large we have,
again to leading order, e6p2x 2/3 and I ( ) e4/3 . Therefore, the generic dominant
asymptotic behavior is just

edom = 1 + O e2/3 .
(5.18)
ereg = 1 +
This tells us that we must simply search for values of k 2 for which the regular solution tends to
zero for large .
We find a discrete spectrum of spin-2 states, the first of which are reported and compared to
Krasnitz values in Table 2. Krasnitz results, which were obtained using a WKB approximation,
are in good agreement with ours. As shown in Fig. 3, the spectrum fits nicely to a quadratic curve.
21 There is only one dominant and one subdominant asymptotic behavior, and the latter is sufficiently suppressed for
large with respect to the former.
20
Table 2
Mass spectrum of spin-2 glueballs in the KS background (m2 < 100), and comparison with Krasnitz WKB results [42].
Krasnitz values have been normalized such that the seventh masses (the highest he calculated) agree
n
Krasnitz
n
m2
n
m2
n
m2
1
1.06
1
1.044
8
21.62
15
68.78
2
2.39
2
2.369
9
26.73
16
77.69
3
4.227
10
32.38
17
87.15
3
4.52
4
6.65
4
6.624
11
38.57
18
97.15
5
9.561
12
45.31
5
9.62
6
13.1
6
13.04
13
52.59
7
17.1
7
17.06
14
60.41
Fig. 3. Mass-squared values of spin-2 states. The solid line represents the fit (5.19).
A least-square fit yields

m2 0.2715n2 + 0.4936n + 0.2969.
As we will find in the next section, the coefficient of the
universality.
(5.19)
n2
term enjoys a certain degree of
5.3. KS 7-scalar system: Setup

Now, we turn to the most difficult system of our paper: the seven coupled scalar fields in the
KS background. All scalars appear to be fully coupled in the bulk, but we follow the insight
gleaned from the moderate UV approximation [11] to decouple a 4 4 set of fields (the glueball sector) from a 3 3 set (the gluinoball sector) to leading order in the UV, as will be seen
in Section 5.4. The system of field equations we consider follows from (2.5) upon changing the
radial coordinate to . One finds

( M)( N ) k 2 e2A8p a = 0,
(5.20)
where we have dropped the tensor indices, and the matrices M and N are defined by
M a b = N a b K a b 2e2x6p ba ,

W a Wb
a
4p
a
N b =e
,
b W
W
K a b = 2e4p G a bc W c .
21
(5.21)
As before, we fix the momentum scale as in (5.15).

Now, the matrices M and N a priori depend on the constants P and 0 . However, there
is a linear field transformation that removes this dependence from (5.20). This implies that
these constants can affect the mass spectrum only by an overall change of the momentum
scale, which is not visible in our effective 5d approach. Starting with the fluctuation vector
a = (x, p, y, , b, h1 , h2 )T , the linear transformation that accomplishes this is a = Ra
with

h1
h2 T
a = x, p, , , y, b,
(5.22)
Pe 0
P e0
and we also rotate the matrices by, e.g., N = RNR 1 . Henceforth, we shall consider the rotated
matrices, dropping primes. The somewhat lengthy expressions for the rotated matrices K and N
(that are now independent of P and 0 as advertised) as well as the sigma-model metric G can
be found in Appendix C.1.
5.4. KS 7-scalar system: Boundary conditions
In this section, we will consider the asymptotic (large- ) and deep-IR (small- ) behavior of
the solutions of (5.20), which are needed to fix the boundary (initial) conditions for the numerical
integration.
Let us start with the large- region. Asymptotically, the matrices and the warp term in (5.20)
can be expanded in powers of e , such that22

N = N (0) + e N (1) + O e2 .
K = K (0) + e K (1) + O e2 ,
(5.23)
In what follows, we can always drop the O(e2 ) terms (the reason for this will become clearer
later on, cf. the discussion below (5.40)). For the sake of brevity, we write the matrices in block
form,

K44 K43
K=
(5.24)
,
K34 K33
and analogously for N . Then, the matrices in (5.23) are
0 0
2
0
3( 1/4)
0 0
(0)
K44 =
2 0
0
0
1
1/4
4
3( 1/4)
0
0
(0)
1
K33 = 0 1/4
2
,
1
(0)
(0)
K43
= K34 = 0,
(5.25)
4
3( 1/4)
(5.26)
1
1/4
22 The coefficients may contain rational functions of , but no exponentials.
(5.27)
22
(0)
N44
1
+1/4
2( 1/4)
3( +1/4)
=
1
+1/4
0
1 0
(0)
N33 = 0 0
2
(0)
N43
and
(0)
= N34
1
4+1/4
2
3( +1/4)
+5/4)
2(
3( +1/4)
4 1
+1/4
2
9( +1/4)
2
3( +1/4)
0
1,
4
3( 1/4)
0
0
0
4
8( 1)
3(
1/4)
8
3( 1/4)
0
0
(1)
K34 = 4( 1) 0
4( 2)
(1)
= K33
= 0,
1
+1/4
2( 1/4)
3( +1/4)
(1)
N44
0
4
2( 2)
4( 1)
3(
+1/4)
4( 1)
9( +1/4)
+1/4
16 1
3( +1/4)
2
+1/4
2( 1)
=
+1/4
2)
2(
+1/4
(1)
= N33 = 0.
0
2( 1) ,
8
3( 1/4)
(1)
N43 =
4 2 5 5/2
(1)
0
(5.29)
4( 1)
3( 1/4)
N34
(5.28)
(5.30)
0
,
(1)
K43 =
4( 2)
= 0,
(1)
K44
8( 1/4)
+1/4
8( 1)( 1/4)
+1/4
8( 2)( 1/4)
+1/4
4
3( +1/4)
(5.31)
(5.32)
(5.33)
9( +1/4)
,
(5.34)
3( +1/4)
0
4
3( +1/4)
4( 1)
3( +1/4)
4( 2)
3(
+1/4)
2( 1)
,
(5.35)
2( 2)
(5.36)
The transformation a a , cf. (5.22), has brought the matrices into this nice block form. We
also need

2
e2x6p = + O e2
(5.37)
3
as well as

31/3
1 2/3
1 + O e2 .
e
e2A8p =
(5.38)
h(0)
4
It is a useful check that the leading order terms of these expressions coincide with the respective
quantities evaluated in the KlebanovTseytlin background [26].
Since (5.38) is exponentially suppressed, the leading order terms of the asymptotic solutions
will be independent of the momentum k. We note that this is just as we assumed in Section 2.3.
In contrast, in the MN system, the asymptotic behaviors of the solutions depend on k [cf. (4.13)].
23
The asymptotic UV solutions, including the leading and some subleading terms, are found by
iteratively solving the equations

N (0) (n) = (n) + e N (1) (n1) ,
(5.39)

1 2/3 (n1)
+ e M (1) (n1) ,
e
M (0) (n) =
(5.40)
4
where = 31/3 k 2 / h(0), and we set (1) = (1) = 0. The solutions (0) are the leading order
terms of the asymptotic solutions. Again, we will drop all O(e2 ) terms in the iteration. The
reason for this is the following. Dropping subleading terms in the initial conditions leads to
systematic errors in the numerical solutions, which must obviously be kept smaller than the
relevant parts of the solutions. This implies that the initial conditions of all behaviors we use
must be given to the same order in e . We will use the mid-point method of Section 3.2, in
which only the subdominant asymptotic behaviors are needed. Now, the leading term of the
weakest subdominant behavior goes like e8/3 , whereas the strongest subdominant behavior
goes like e . Thus, they differ by e5/3 and we can drop O(e2 ) corrections.23
The somewhat lengthy expressions of the asymptotic solutions are deferred to Appendix C.2.
Now, let us consider the small- region. In order to find the independent behaviors for small ,
we expand the matrices in (5.20) about = 0 and make an ansatz of the form

a( ) = q a0 + a1 + 2 a2 + 3 a3 + ,
(5.41)
with some undetermined number q, and constant vectors an , n = 0, 1, 2, . . . . These are then
determined recursively from the equation of motion (using computer algebra). It turns out that,
of the 14 solutions, there exist four solutions with q = 0, one with q = 1, three with q = 2,
four with q = 1, and one each with q = 2 and q = 3. Thus, there exist eight solutions,
whose components are finite at = 0, and six with singular component behavior. However,
the regularity condition (2.10) tells us that one of the solutions with q = 0 is, in fact, singular.
Therefore, we arrive at precisely seven regular and seven singular solutions. The independent
small- behaviors are listed in Appendix C.3.
5.5. KS spin-0 spectrum
Analogously to the MN case, instead of 7 coupled second-order ODEs, it is more convenient
to solve 14 coupled first-order ODEs. Thus, let us rewrite (5.20) as

a
a
N
1
.
=
(5.42)
b
b
k 2 e2A8p M
We apply the midpoint determinant method as outlined in Section 3.2. That is, we compute
the 14 14 matrix

areg asub
=
(5.43)
breg bsub =mid
as a function of k 2 and look for zero crossings of det . A rough plot of det is shown in
Fig. 4. We verify that there are no zero crossings for positive k 2 . The zero crossings of det
23 The same argument holds for the method of Section 3.1, where only the dominant solutions are needed. The weakest
dominant solution goes like e/3 , the strongest one like e4/3 .
24
Fig. 4. det as a function of k 2 . For clarity, we have taken the 14th root of the absolute value of the determinant (leaving
its sign untouched). The plot shows clearly that there are no zero crossings for positive k 2 . The inset shows the zeros of
det for the first three spin-0 states.
Fig. 5. Here we plot det in the range k 2 [5.8 . . . 5.3]. The inset zooms in on the region around k 2 = 5.63, where
there are two zeros.
can be found by zooming in on a particular region of the plot, as shown, for example, in the
inset. As the determinant itself changes over many orders of magnitude, it is useful to plot
| det |1/14 sign(det )| det |1/14 instead. The sharp turns in the inset are merely artifacts
of this, as is apparent from Fig. 5, where det itself is displayed without the 14th root, and there
are no sharp turns.
We have calculated all mass values up to m2 = 600. The first few are listed in Table 3; for
a more extended list see Table 5 in Appendix D. Our results exhibit two interesting features.
Table 3
Low-lying KlebanovStrassler spin-0 mass states, extracted from zero-crossings
Appendix D
m2
0.185
0.428
0.835
1.28
2.61
3.32
3.54
4.12
5.35
5.63
5.63
6.59
25
in Fig. 4. More values are given in

1.63
4.18
6.66
1.94
4.43
6.77
2.34
4.43
7.14
Fig. 6. On the left, we see that for high m2 , it is easy to identify the periodicity empirically as 7. On the right, we
have extracted one tower (i.e., every 7th point) from the full spectrum, enumerated states in that tower by a parameter
nt (the excitation number) and fit those points to a quadratic curve starting with the third point. As shown in the inset,
m2 n2t is not a good description for very low nt , unlike in the spin-2 case.
First, it appears that several mass values are nearly degenerate, such as the two values about
m2 4.43 and two values about m2 5.63. The origin of this in our calculation can be seen
in Fig. 5: The det curve barely seems to touch the k 2 -axis, and only zooming in reveals that
there are two crossings. As we do not have a dynamical explanation for this, it is possible that the
near degeneracy is accidental, or it may indicate a multiplet originating from a weakly broken
symmetry of the low-energy effective theory.
The second interesting feature is the appearance, for large masses, of a periodic pattern of
period 7 in the consecutive spectrum excitation number n. On the left-hand side of Fig. 6, this
pattern is easily visible. Hence, we split the spectrum into 7 towers of mass states, one of which
is shown on the right-hand side of Fig. 6. From now on, we distinguish excitation numbers of
individual towers, denoted by nt , from the excitation number n of the whole spin-0 spectrum.
Like the spin-2 spectrum, the seven spin-0 spectra also exhibit quadratic dependence of m2 on
excitation number nt . Leaving out the first two values of each tower (i.e., nt = 1, 2) least square
fits of the large m2 behavior (using mass values up to m2 = 600) yield
m2 0.271n2t + 0.774nt + 0.562,
m2 0.270n2t + 0.928nt 0.430,
m2 0.275n2t + 0.769nt + 1.921,
m2 0.272n2t + 1.017nt + 1.023,
m2 0.272n2t + 1.119nt + 0.337,
26
m2 0.271n2t + 1.150nt + 0.648,

m2 0.273n2t + 1.082nt + 2.544.
(5.44)
The last digit of the leading coefficients (i.e., the coefficients of the n2t terms) can vary slightly
if one performs the fits with more than two states from each tower left out. Still, it is obvious
that within some uncertainties the leading coefficients all coincide rather well, both amongst
each other and also with the coefficient of the spin-2 tower (5.19), leading to roughly universal
behavior at large nt .
5.6. KS 7-scalar system: Quadratic spectra
If one plots the experimental values for squared meson masses m2 (e.g., for the first few resonances of the meson) against their consecutive number n, the data obeys m2 n to good
accuracy.24 Since m2 n is the relation that follows from a linear quark potential, this is sometimes called linear confinement. Karch, Katz, Son and Stephanov (KKSS) [19] emphasized
that there are no known supergravity fluctuations that reproduce this experimental fact about
mesons from AdS/CFT.25
In fact, the claim of KKSS was more specific (p. 13): for all known supergravity backgrounds
the fluctuation spectra will be quadratic for large n, that is, m2 n2 . It has even been claimed
that this is a general feature of the supergravity approximation. However, even if this is true for
large n, it is a priori not clear when this large n behavior should set in, given the complexity
of fluctuations around the KlebanovStrassler solution in this paper, and the claim says nothing
about the small n spectrum.
Although we cannot be absolutely sure from our analysis, the KKSS claim that m2 n2 seems
resoundingly apparent in Fig. 6 and the fits (5.44), already for n 3. (To be precise, this is nt 3
in each tower; for the overall consecutive number this means n 15.) Since the KS theory is in
any case not QCD, the phenomenological question raised in the first paragraph is not directly
relevant here. On the other hand, it would be interesting to know if there are models where an
m2 n relation holds for low spins (the metric-scalar model given in [19] seems promising, but
so far has not been shown to be the solution of any known theory).
Of course, one would expect that much of the low-energy dynamics of the theory would be
determined by the lowest-lying mass states, and they have a richer structure than the large-n
asymptoticsa glimpse of this can be seen in the inset on the right-hand side of Fig. 6. Since
the KS theory incorporates spontaneous breaking of chiral symmetry, this should have left its
mark on our spectrum. We do observe some interesting patterns, but we have not been able to
link them conclusively to symmetry breaking. This deserves further study.
As alluded to in the introduction, operator mixing is even more subtle here than in standard AdS/CFT. We have computed the spin-0 mass spectrum, but we have not said what the
mass states are composed of in terms of dual gauge theory operators. Some of the problems
of identifying the mass states were already addressed in Section 2.3, i.e., the ambiguity in the
normalization of the dominant asymptotic solutions and the freedom to add to a given dominant
24 We omit the subscript t on n from now on.
t
25 Of course, much work has been done to go beyond the supergravity approximation, in particular for states of high
spin, where m2 n behavior is seen at least for some range of masses. Here the issue is what the spectrum looks like in
the supergravity approximation, and for low spin. As usual, by supergravity approximation we understand gs N 1
and gs M 1.
27
solution multiples of asymptotic solutions of equal or weaker strength. Thus we do not attempt
to give a further interpretation of the mass states. We just content ourselves by noticing that there
is actually a quantity that is easy to obtain with our method, namely the coefficients d,i from
(2.28). These can be determined straightforwardly by solving (3.8) at the various mass values.
This gives information on the composition of the cavity mode (that we called a in Section 2)
corresponding to a mass state in terms of our basis of subdominant solutions. However, as the
interpretation in terms of operator mixing is not clear to us, we refrain from determining these
coefficients explicitly here. Moreover, also the quantities Z,i , defined in (2.33), could be determined in principle. They correspond to the residues of the mass poles if N , introduced in (2.32),
is really one, as in the case of asymptotic AdS spaces (see Appendix A).
5.7. Comparison to hardwall approximations
In this section, we would like to give some arguments supporting our conviction stated in
the introduction that the detailed form of the spectrum depends crucially on the details of the
IR dynamics in the gauge theory. Holographically, this means that it is important to use the full
KlebanovStrassler background to determine the low lying mass eigenvalues. To illustrate this,
we compare our spectrum with a naive toy hardwall model. In particular, one might hope to find
the right spectrum by using the KlebanovTseytlin approximation of the KS background. This
is roughly equivalent to cutting off the KS background at some value IR of order one (the two
solutions differ strongly only for 0.8). In order to use the determinant method described in
Section 3.1, one would like to impose seven independent initial conditions at IR and then solve
the equations numerically from the IR towards the UV. One again demands that there is a linear
combination of those numerical solutions that does not contain any dominant solutions adom
when expanded in a basis of asymptotic solutions. Lacking any guiding principle for choosing
these initial conditions, we arbitrarily impose that all components of the seven scalar fields should
vanish at IR , whereas for each initial condition one of the first derivatives are taken to be unity.
This gives seven independent initial conditions. The resulting spectrum is contained in Table 4
for several values of IR , together with the correct mass values. Obviously, the toy spectra
depend on the value of IR , but they also differ strongly from the correct values. We believe that
Table 4
Comparison of our mass values with a toy hardwall model as described in the text (only values up to m2 = 5 are included).
This demonstrates that the spectrum indeed strongly depends on the IR dynamics
Values from Table 3
IR = 0.5
IR = 0.75
IR = 1
0.19
0.42
0.83
1.29
1.63
1.94
2.35
2.61
3.33
3.53
4.13
4.17
4.43(2)
0.27
0.56
0.86
1.39
1.78
2.32
2.43
3.26
3.58
4.06
4.30
4.54
4.71
0.30
0.63
1.03
1.62
1.89
2.42
2.56
3.39
3.80
4.22
4.60
4.81
0.34
0.71
1.14
1.83
2.03
2.57
2.75
3.70
4.09
4.40
28
this holds more generally, even if one had a different set of initial conditions than these, whether
or not they are physically motivated.
To summarize, we do not see any way of implementing a hard-wall approximation that reproduces the correct spectrum of Table 3.
6. Outlook
We have computed spin-0 and spin-2 mass spectra for the MaldacenaNuez (wrapped D5brane) and KlebanovStrassler (warped deformed conifold) supergravity duals. Although the
spectra are fairly complicated, there are some simple features. For example, for large excitation
number n, the spin-0 states in the KS theory organize into 7 towers with m2 n2 , in agreement with the claims of [19]. (Large is not terribly large; around n 3.) For low excitation
number n, there is a rich structure of mass values, which exhibits interesting patterns of near
degeneracy.
For the MaldacenaNuez background, our numerical results confirm the rough features of
the analytic spectra we derived in [11] and improved upon here; the discrete mass spectrum has
an upper bound, which is quite different from the KS case.
The situation with respect to other studies of glueball spectra in these backgrounds, based on
various truncations, has not changed from [11]. The one result we have been able to compare
our results to is the set of mass states for spin-2 glueballs in KS, where the first few values were
computed in the WKB approximation by Krasnitz in [42], and are roughly correct, as in Table 2.
It would be very important to understand the details of mixing in this context; without further
information along the lines we have suggested, we cannot determine the composition of the mass
states we find. On the good side, the spectrum we have found is unique, in that the variables used
to find it are gauge-invariant, and we impose the boundary conditions one really wants to impose,
rather than approximations thereof.
These results and methods could straightforwardly be applied to other gauge/gravity dual
pairs. Apart from other N = 1 dualities, it would be interesting to apply these methods also in
N = 0 gauge theory. There, the existence of a simple superpotential W , which greatly simplified
our task, is not guaranteed but occurs, e.g., for N = 0 backgrounds with pseudo-Killing spinors
(cf. for example [23]). Here is a concrete question: do the meson spectra computed in other
backgrounds contain glueball admixtures from couplings in the DBI action?
Independently of supersymmetry breaking, including probe branes in KlebanovStrassler la
[20,21] would be interesting to study using our methods.
Finally, the greatest challenge for holography in presently known confining backgrounds is
to compute correlators from the KlebanovStrassler background. Some progress was made in
[7,42,4446]. We considered the Krasnitz (extreme high-energy) limit in [11] and found some
analytical solutions. However, those solutions were checked to be valid around k 2 106 , whereas
in this paper, we were interested in IR physics, and thus the Krasnitz approximation was not
available to us. In fact, we only considered solutions up to k 2 2000, and going much higher
presents a numerical difficulty, as one needs to go increasingly far in the UV for the solutions to
become asymptotic, so that the determinant method is applicable. This is a superable challenge
for connecting the high-energy and low-energy regimes, which might ultimately be the most
useful aspect of holography for confining gauge theories. We hope to return to these issues in
future publications.
29
Acknowledgements
It is a pleasure to thank Massimo Bianchi, Andreas Karch, Emanuel Katz, Albion Lawrence,
Scott Noble, Carlos Nuez, Henning Samtleben, David Tong and Amos Yarom for helpful discussions and comments. This work is supported in part by the European Communitys Human
Potential Program under contract MRTN-CT-2004-005104 Constituents, fundamental forces
and symmetries of the universe. This research was supported in part by the National Science
Foundation under Grant No. PHY99-07949. The work of M.B. is supported by European Communitys Human Potential Program under contract MRTN-CT-2004-512194, The European
Superstring Theory Network. He would like to thank the Galileo Galilei Institute in Florence for
hospitality. The work of M.H. is supported by the German Research Foundation (DFG) within the
Emmy Noether-Program (grant number: HA 3448/3-1). Both M.B. and M.H would like to thank
the KITP in Santa Barbara for hospitality during the program String Phenomenology. The
work of W.M. is supported in part by the Italian Ministry of Education and Research (MIUR),
project 2005-023102.
Appendix A. Some 2-point functions in AdS/CFT
In this section, we review how the 2-point functions for some known cases of AdS/CFT arise
as sums over the spectrum of bulk eigenfunctions, using expressions developed in the main text.
A.1. AdS bulk
Let us start with a set of n free massive scalar fields on (d + 1)-dimensional AdS bulk space.
Using the language of Section 2, we will derive the 2-point function as a sum over the spectrum
of bulk eigenfunctions. (For pure AdS the spectrum is continuous, so the sum is in fact an
integral. This will be different in the next subsection.) When the smoke has cleared (in (A.14)),
the connection to the standard computation in terms of the IR-regular Bessel function K (kz)
will be evident.
The bulk equation of motion (2.5) (in d-momentum space) is

d 1
m2
2
2
z 2 k = 0,
z
(A.1)
z
z
where z = er is the radial coordinate. We have omitted the field indices, and m2 is to be understood as a diagonal matrix with entries m2i , i = 1, . . . , n. The conformal dimensions of the
operators Oi dual to the components of are related to the mass parameters mi by26

d2
d
i = + i , i =
(A.2)
+ m2i .
2
4
Small values of z give the asymptotic UV region, whereas large z describe the bulk interior (IR
region).
26 In general one needs to consider = d , but for the purposes of this discussion we restrict to the upper sign.
i
i
2
For simplicity, we also consider generic (non-integer) values of i .
30
Conventionally normalized asymptotic solutions of (A.1) are27

i
k
dom ai (z, k) = ia (1 i )
zd/2 Ii (kz) = ia zd/2i + ,
(A.3)
2
i
k
a
a
zd/2 Ii (kz) = ia zd/2+i + ,
sub i (z, k) = i (1 + i )
(A.4)
2
where the I are modified Bessel functions, and after the second equal signs we indicated just
the respective leading behaviors. The powers of k in front of the solutions are necessary in order to make the leading terms k-independent, while the remaining coefficients are conventional
normalizations.
For asymptotically AdS bulk backgrounds, (A.1) still holds in the asymptotic region. Therefore, the leading behaviors of the asymptotic solutions remain as in (A.3) and (A.4),
dom ai (z, k) = ia zd/2i + ,
sub ai (z, k) = ia zd/2+i + .
(A.5)
The general form of the matrix Zij (2.30) in asymptotically AdS spaces follows easily. One finds
Zij = 2i ij .
(A.6)
A general result of holographic renormalization [3638] is that (the non-local part of) the
exact one-point function Oi exact is
(A.2)
Oi exact = 2i di = (2i d)di .
(A.7)
Comparing this with (2.24), one finds

Yij = 2i ij = Zij ,
(A.8)
which, together with (2.32), implies N = 1.

Returning to the pure AdS bulk, (A.1) admits a continuous spectrum of regular and subdominant solutions for k 2 = 2 , > 0. The eigenfunctions are given by28
a
i
(z) = ia zd/2 Ji (z)
(A.9)
and satisfy the orthogonality relation
dz (d2)
i (z) j (z) = ( )ij .
z
z
(A.10)
Considering the small-z behavior of (A.9), one can read off the response coefficients di,j ,
i
di,j = ij
(A.11)
.
2
(1 + i )
Hence, after using (2.33) and N = 1, one obtains for (2.34)

22(1i )
Oi (k)Oj (k) = ij
(i )2

0

2i +1
(1 i ) k 2i
d 2
= ij 2
.
(i )
2
k + 2
(A.12)
27 We have reinstated the field index a.

28 The generic label for the eigenfunctions used in the main text [cf. (2.14)] is replaced here by two indices, = k 2
and i = 1, . . . , n. As before, the upper index a is the vector component index.
31
Notice that the second equality holds only after analytic continuation, because the integral does
not exist if i 0. This is equivalent to adding an infinite contact term to the integral over the
spectrum. For example, if 0 < i < 1, we rewrite the integrand as
2i 1
2i +1
2i 1
2
=
k
k 2 + 2
k 2 + 2
(A.13)
and add a counter term that cancels the first term on the right-hand side.
The finite result (A.12) can also be obtained in the usual way, i.e., by considering a regular
solution of (A.1),
ia (z, k) = ia
i
1i
2d
i +1 d/2
k
d/2
a 2
z Ki (kz) = i d
d 2
z Ji (z), (A.14)
(i ) 2
(i )
k + 2
0
with the help of

reading off the relation between the response and source coefficients, d and d,
(1 )
(A.15)
x (1 + )
and then using (A.7). It is interesting to note that the small-z behavior of the integrand in (A.14)
is only subdominant, while the expression in terms of Ki contains also a dominant piece. The
explanation of this apparent discrepancy is the UV part of the spectrum (large ), i.e., one cannot
find some sufficiently small z such that z is also small for all values of . Hence, loosely, the
UV-part of the spectrum generates the dominant behavior.
K (x) = 21 ()x (1 + ) 21
A.2. Active scalar in GPPZ flow

As a second example, let us take the active scalar in the GPPZ flow [47]. Its gauge invariant
bulk equation of motion reads [31,32]

3u
k2
2
u(1 u)u + (3u 2)u +
(A.16)
1+
= 0,
4(1 u)
4
where the radial coordinate u is defined by u = 1 e2r , and the warp factor is e2A = u/(1 u).
Eq. (A.16) admits a discrete spectrum of regular and subdominant eigenfunctions, with mass
squares
m2n = 4n(n + 1),
n = 1, 2, 3, . . . .
The normalized eigenfunctions are

2(2n + 1)
d
n =
(1 u)3/2
Pn (2u 1),
n(n + 1)
du
where Pn are Legendre polynomials.29 One easily finds the response coefficients

dn = 2n(n + 1)(2n + 1).
(A.17)
(A.18)
(A.19)
29 As in [31,32], regular and subdominant solutions of (A.16) are given by Jacobi polynomials P(1,1) (z), which are
n1
proportional to dPn (z)/dz.
32
Thus, using = 1 (we have a = 3 operator) and (A.8) (which holds for asymptotically AdS
spaces), we obtain for the 2-point function (2.34)

8n(n + 1)(2n + 1)
+ c.t.
O(k)O(k) =
k 2 + 4n(n + 1)
(A.20)
n=1
Clearly, the sum in (A.20) does not converge, so that there are again infinite contact terms. It is
instructive to compare (A.20) with the finite result from holographic renormalization [48,49],

k2
3 + 1 k2
3 1 k2
O(k)O(k) =
(A.21)
+
(1) (2) ,
2
2
2
where (z) = [ln (z)] . Using the identity

1
1
(x) (y) =
,
y +n x +n
(A.22)
n=0
we obtain from (A.21)
k4
2n + 1
O(k)O(k) =
2
n(n + 1)[k 2 + 4n(n + 1)]
n=1

8n(n + 1)(2n + 1)
2 2n + 1
2(2n
+
1)
+
k
.
=
2n(n + 1)
k 2 + 4n(n + 1)
(A.23)
n=1
Again, we see that holographic renormalization makes a precise choice for the contact terms and,
therefore, picks a renormalization scheme and scale. Indeed, the finite result (A.21) satisfies the
renormalization conditions

d
O(k)O(k) k 2 =0 = 0,
(A.24)
O(k)O(k) k 2 =0 = 0.
2
dk
Appendix B. Matrices for the MN background
The entries of the matrices M and N of Eq. (4.7) for the regular MN background (c = 0) are

M11 = 8 2 4 + 1 + e4 64 3 + 112 2 48 + 2

+ e8 256 4 896 3 + 248 2 + 108 17

+ e12 1536 4 736 2 + 28 + e16 256 4 + 896 3 + 248 2 108 17

+ e20 64 3 + 112 2 + 48 + 2 + e24 8 2 + 4 + 1

4

/ 1 e4 1 e4 (4 + 1) 4 1 + e4 ,

M12 = (3 1) + e4 16 3 48 2 + 12 + 2 + e8 96 3 30

+ e12 16 3 + 48 2 + 12 2 + e16 (3 + 1) 4e2
3

/ 1 e4 1 e4 (4 + 1) 4 1 + e4 ,

M21 = 1 e4 (4 + 1) 4 1 + e4 ( 1) + e4 6 + e8 ( + 1) 4e2

5
/ 1 e4 ,
33

M22 = (2 1) + e4 32 3 128 2 + 56 2

+ e8 256 4 + 960 3 256 2 118 + 17

+ e12 1536 4 + 768 2 28

+ e16 256 4 960 3 256 2 + 118 + 17

+ e20 32 3 128 2 56 2 + e24 (2 1)

4

/ 1 e4 1 e4 (4 + 1) 4 1 + e4 ,
2

M33 = 8 1 e4 / 1 e4 (4 + 1) 4 1 + e4 ,
M13 = M23 = M31 = M32 = 0,
and
(B.1)

N11 = 1 + e4 16 2 32 + 4 + e8 96 2 10 + e12 16 2 + 32 + 4

4
+ e16 / 1 e4 ,
3

N12 = ( 1) + e4 6 + e8 ( + 1) 4e2 / 1 e4 ,

N21 = (3 1) + e4 16 3 48 2 + 12 + 2 + e8 96 3 30

5

+ e12 16 3 + 48 2 + 12 2 + e16 (3 + 1) 4e2 / 1 e4 ,

N22 = (2 1) + e4 16 2 + 28 4 + e8 96 2 + 10

4

+ e12 16 2 28 4 + e16 (2 1) / 1 e4 ,
N13 = N23 = N31 = N32 = N33 = 0.
(B.2)
Appendix C. Bulky KS stuff

C.1. KS matrices
We present here the explicit expressions for the 7 7 matrices appearing in Section 5.3. To
shorten the formulae, a number of abbreviations will be used. First,
c = cosh y = coth ,
s = sinh y = (sinh )1 ,
(C.1)
where y denotes the background field of Section 5.1. Second, we introduce

B1 = c 1,
B2 = s 2 c,
(C.2)
and
A1 = h( )(4sB2 )1/3 = h( ) sinh (2 sinh 2 4 )1/3 ,

2
1
sB1 B2 .
A2 = A1 cB2
3
2
(C.3)
(C.4)
Let us consider the behavior for small and large of A1 and A2 . As h(0) is a finite, positive
constant, one obtains
1
A1 (0) = 31/3 h(0),
2
4
A2 (0) = A1 (0).
3
(C.5)
34
For large , starting from

h( ) 3e
4/3

1
,
(C.6)
one obtains

3
1
,
A1 ( ) e
2
4

3
1
A2 ( ) e +
.
2
4
(C.7)
With the abbreviations (C.1)(C.4), the (rotated) matrices K a b = 2e4p G a bc W c and N a b are
given by
2(1 + 2cB2 )
0
K =
2sB1
2s( + 2B2 )
s
2A1 B2
s B1
2A
1 B2
s2
2A1 B2
2(A2 +cA1 B2 )
A1 B2
2cB2 + 1
2s( + 2B2 )
2s
s
A1 B2
s 2 B1
A1 B2
s2
A1 B2
As1 B 2
A1sB2
sB1
2(A2 +cA1 B2 )
A1 B2
2(A2 +cA1 B2 )
A1 B2
2s
s( + 2B2 )
2(1 + 2cB2 )
(C.8)
1 B2 )
2c(A2 +cA
A2
2cA
3A2
2(1+2cB2 )(A2 +cA1 B2 )
A2
0
N =
2s(A
+cA1 B2 )
2

A2
2sB1 (A2 +cA1 B2 )
A2
2s( +2B2 )(A2 +cA1 B2 )
A2
1
4cA
A2
cs
2A2
s
6A2 B2
s(1+2cB2 )
2A2
(1 + 2cB2 )
1
4sA
A2
s2
2A2
1
sB1 +2cA
A2
2)
4A1 (1+2cB
A2
4sAA12B1
4sA1 ( +2B2 )
A2
cs2AB21
sA1
3A
2
s B1
6A
2 B2
2A B B
1 1 2
A2
2A2 (1+2cB2 )+s 2 A1 B2 ( +2B2 )
A2
The matrix M is given by
s 2 B1
2A2
s 2 ( +2B2 )
2A2
1 B2 )
s(A2 +cA
A2
1 B2 (1+2cB2 )
2sA2 ( +2B2 )+sA
A2
0
s 2 A1 B2
A2 c
sB1
s( + 2B2 )
2 B (1+2cB )+4csA
1
2
2
2A2
0
3
s2AB21
s3B2
2A21
s 3 B1 ( +2B2 )+2A2 (2s 2 +1)

2A2
cs 2
2A2
s 2 (1+2cB2 )
2A2
.
0
2A2
3
s B1
2A2 + 1
s2
6A2 B2
3 ( +2B
2A2
2)
(C.9)
M = N K +
35
4
I,
3B2
(C.10)
where I denotes the 7 7 unit matrix.

Finally, we also need the sigma-model metric for the rotated fluctuation fields. It transforms
as G = (R 1 )T GR 1 , where R is the linear transformation matrix that leads to (5.22), and the
superscript T denotes the transpose. Explicitly, we find
1 0
0 6
0 0

G = 0 0
0 0
0 0
0 0
0
0
0
0
0
0
1
4
1
2
1 2 0 2x
2P e
1 2 0 2x
2P e
cosh(2y)
0
0
0
1 2 0 2x
2P e
sinh(2y)
1 2 0 2x
sinh(2y)
2P e
0
,
0
0
1 2 0 2x
2P e
cosh(2y)
(C.11)
where for x and y one should substitute the respective background solutions.
C.2. KS 7-scalar system: Asymptotic solutions
In this appendix, we list as reference the asymptotic solutions of KS the 7-scalar system. They
are found by iteratively solving (5.39) and (5.40).
For convenience, we abbreviate
=
31/3 2
k .
h(0)
(C.12)
Moreover, 03 and 04 denote 3 and 4 zero vector entries, respectively.

The dominant asymptotic (large ) solutions, up to and including terms of order e/3 (why
exactly e/3 is explained in Section 5.4), are
12
6(5 + 4 )
04
(9 + 4 )
4
e4/3
9
24 /3
1
e2/3 6(5 + 4 ) +
e
12 +
32(4 + 1)
4 + 1
1
4 + 1
0
0
2
03
03
24 2 48 63/2
04
8 + 9
1
27(4 + 5)
27 2
e/3
+
24 2 + 48 + 63/2
,
1
256(4 + 1)
8(4 + 1)
0
2
03
(C.13)
36
04
04
04
2/3
0 9
8 2 30 + 45
9 2

2

e +
e/3
e/3
,
+ 4 2
256
1 32
2 2
6 2 39 + 243/2
0
6 2 + 6 81
1
1 2
03
(C.14)
4 + 13
36 + 63
2 7/2
8 18

e2/3
12 9
24 2 + 12 279/2
4 + 1
32
72 + 42
0
0
3
04
2
e/3
64 104 6
,
4 + 1 120 2 246 99
24 2 + 174 + 99
1/2
1/6
1,
03
(C.15)
(C.16)
03
1/2
1/6
1/4 ,
4 + 1
0
03
04
2 + 1
e/3
3 + 3/2
(C.17)
(C.18)
9/4
and
04
4
e/3
.
9
(C.19)
3
The subdominant asymptotic (large ) solutions, up to and including terms of order e8/3
(why exactly e8/3 is explained in Section 5.4), are
37
32 29
04
04
1
9
e2 6 (16 + 37)
4 + 11
1
5/3
e
e
7(16 7)
+
+
16
2 2 + 12
20
0
2
1
2 + + 85/4
03
27 2 7/3
6400
04
300 2 + 1625 + 627
33 1131 13 649/10
3
3
2
40 (1600 + 2440 35 920 27 383)
80 3
3(35 200 3 27 760 2 316 608 + 42 739)
4(800 3 + 4240 2 15 228 33 961)
+
e8/3 3(169 600 3 + 219 680 2 705 996 67 973) ,
16 000(4 + 1)
0
03
(C.20)
04
0
1
e5/3
+
+ 5/8
04
1
+ 1/8
3 2 7/3
640
3
13/6
2
1
+e
03
04
195 + 2989/12
1
2
50 (3150 + 1505 5591)
1
100
(11 700 2 5260 28 993)
3(40 2 + 682 + 69)

(4 + 11/2)(140 + 57)
+
e8/3 3(620 2 709 117) ,
400(4 + 1)
0
03
1
80 2 32 291
1
40 2 356 5807/6
4/3
2
e
e
6 3 +
1680 2 + 2732 1634
800
4 + 9
50(4 15)(4 + 9)
03
03
(C.21)
38
04
2000/3
3 7/3
400 2 + 260 547

250
400 2 + 2260 297
34 560 3 111 024 2 + 259 556 497 927/2

11 520 3 104 688 2 1 489 028/3 2 917 225/6
+
e8/3
470 880 3 + 1 839 672 2 135 995 766 688
, (C.22)
204 800
3(57 600 3 36 000 2 1 110 600 1 043 879)
+
03
(12 2 51 23)
3
1 (144 2 56 + 53)
24
e4/3
3
2
e
12
+
2(56 2 + 17 4)
20(4 + 1)
4 + 1
4(4 + 1)
5(4 + 1)2
03
03
04
50
3
+
e7/3
80 2 74 + 49
25(4 + 1)
80 2 + 226 51
1152 3 + 2720 2 + 7122 + 1763

384 3 992 2 /3 + 650 + 361/3
9 2
8/3
+
e
8256 3 + 10 444 2 1392 4169/4 ,
10 240(4 + 1)
3(320 2 + 200 343)(4 + 1)
03
80 2 + 144 + 5
4 + 1/5
8 (20 2 + 36 + 11)
2 + 23/30
3
e2
e8/3 (80 2 + 144 + 5) ,

4 1/5 +
4 + 1
160(4 + 1)
0
0
03
03
04
1/2
e7/3
3
50 (5 + 4)
(C.23)
(C.24)
(C.25)
3
100
(10 17)
and
3(160 2 172 + 1)
(160 2 + 308 + 121)
e8/3
6(260 2 107 16) .
30(4 + 1)
450(4 + 1)
03
(C.26)
39
C.3. KS 7-scalar system: Small- behavior

In this appendix, we provide the independent small- behaviors. In the calculations, only the
regular solutions are needed. We include subleading terms up to and including order 2 , as they
are needed to avoid systematic errors in the initial conditions. For completeness, we also list the
singular solutions. In the following, we abbreviate
1
A1 = A1 (0) = 31/3 h(0).
2
(C.27)
The seven regular small- solutions are
4
2
4 + O ,
16A1
(4/3)
4
2
0 +O ,
0
(4/3)
13
3/2
(32A1 /3)

+ O 4 ,
2
20
17
0
(16A1 )

0
3k 2 /20 + 9/(80A1 )
0
3k 2 /40 + 9/(160A1 )

0
1

+ 2
+ O 4 ,
0
0

0
3k 2 /20 + 11/(80A1 )

0
1
(1/6)

0
k 2 /60 + 1/(120A1 )
0
k 2 /120 + 1/(240A1 )

0
0

2
1 + 2
+ O 4 ,
k /6 + 1/(6A1 )

0
k /60 + 19/(120A1 )

0
0
40
3
2k 2 /5 3/(8A1 )
1
13k 2 /60 + 1/(16A1 )
0
0

0 + 2
+ O 4 ,
0
0
k 2 /10 + 3/(8A1 )
0
0
0

2
1
k /5 1/10 + 31/(120A1 )
0
k 2 /60 1/20 + 31/(240A1 )

0
0

0 + 2
+ O 4 .
2/(3A1 )

2
3k /10 9/10 + 109/(120A1 )

(C.28)
The seven singular small- solutions are

1
0
0
0
0
1 8A1 /3
1
2 + O( ),
+ O( ),

0
2
0
0
1
0
0
0
0
0
4A1 /9
1
1
12 + O( ),
1 + O( ),
1
0
4A1
0
2
1 4A1 k /3 + 4A1 /9 + 1/6

1
0 +
+ O( ),
0

0
0
0
1
0
1
0
1
12 + O( ),
6

0
0
41
1
0
1/2
1
1
0 +
2
4A1 k 22A1 /5 2/3
0
0
0
2
16A1 k /3 + 20A1 /3 + 2/9
+ O( ).
+
0
(C.29)
0
Appendix D. KS spin-0 spectrum
Table 5 gives a somewhat more extensive collection of the spectrum we have found (we computed the spectrum up to m2 = 600).
Appendix E. Numerics
In this appendix, we give a few details on how we computed the KS mass spectrum and how
sensitive this spectrum is to numerical uncertainties. (Not very sensitive at all, as we now argue.)
A quick summary is given in Table 6. The main program parameters are IR , UV , mid and
nimposed . Let us begin by explaining what they are.
Table 5
KlebanovStrassler spin-0 spectrum, first 75 values
n
m2
n
m2
1
0.185
16
5.63
17
5.63
2
0.428
3
0.835
18
6.59
19
6.66
4
1.28
5
1.63
20
6.77
21
7.14
6
1.94
7
2.34
22
8.08
23
8.25
8
2.61
9
3.32
24
8.57
25
9.54
10
3.54
11
4.12
26
9.62
27
9.72
12
4.18
13
4.43
28
10.40
29
11.32
14
4.43
15
5.36
30
11.38
n
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
m2
12.09
12.99
13.02
13.31
14.23
15.03
15.09
16.16
16.89
17.03
17.44
18.61
19.22
19.40
20.79
n
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
m2
21.33
21.58
22.10
23.53
23.95
24.24
25.94
26.32
26.67
27.30
28.95
29.25
29.62
31.57
31.93
n
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
m2
32.30
33.04
34.82
35.21
35.54
37.65
38.17
38.47
39.32
41.15
41.79
42.01
44.22
45.01
45.19
42
Table 6
A few sample mass values m2 and their variation with some of the program parameters. nimposed is the number of nonadaptive, imposed steps, within which we allow for adaptivity. xx means the plot lacks detail to resolve further digits.
The conclusion is that even under strong changes of program parameters, roughly three significant digits remain stable
IR
UV
mid
nimposed
m2
IR
UV
mid
nimposed
m2
0.001
0.2
0.001
0.001
0.001
20
20
11
25
20
1
1
1
5
1
500
500
500
500
1
0.18500
0.18501
0.18500
0.18500
0.1850x
0.001
0.2
0.001
0.001
0.001
20
20
11
25
20
1
1
1
5
1
500
500
500
500
1
4.12131
4.12129
4.12131
4.12131
4.12131
0.001
0.2
0.001
0.001
0.001
20
20
11
25
20
1
1
1
5
1
500
500
500
500
1
1.28107
1.28121
1.28107
1.28107
1.285xx
0.001
0.2
0.001
0.001
0.001
20
20
11
25
20
1
1
1
5
1
500
500
500
500
1
6.59516
6.59519
6.59516
6.59516
6.5951x
The first two are self-explanatory: the actual radial interval of the KlebanovStrassler background is = [0, ], but to put the equations on a computer we need to impose some cutoffs.30
Thus, IR is an IR cutoff and UV is a UV cutoff. The parameter mid is the midpoint for the midpoint determinant method (see Section 3.2), i.e., the position in where we compute the matrix
ij . We will explain the parameter nimposed in a moment.
To integrate the 14 first-order ODEs between IR and UV , starting from analytically known
boundary conditions given in previous appendices, we use a standard adaptive RungeKutta
4th5th-order solver,31 implemented in Maple, Fortran and C. For purposes of checking
individual mass values, the Maple code is sufficient. For larger scans of parameter space, the
Fortran and C codes were useful. For a typical run of 1000 values of k 2 with double precision,
e.g., to generate the plots in Fig. 4, the C code takes about one hour on a 3 GHz Pentium 4 PC
running Linux.
Many of the fields vary quickly in this system, and it turns out that the stepsize adjustment in
typical adaptive routines does not allow sufficiently fast change, so we imposed a certain number
of steps nimposed that are not adaptive, and then allowed adaptivity within those. Those steps were
exponentially spaced,

i+1 =
UV
IR
1
nimposed
i ,
i = 1, . . . , nimposed .
(E.1)
So nimposed is a program parameter that when set to 1 reduces the numerical integration to standard RungeKutta 4th5th order over the entire range [IR , UV ].
We performed some more extensive tests, but a few sample tests are shown in Table 6, along
with the variation in the answer, the mass value m2 .
So far, this all referred to the midpoint determinant method. We also tried some other approaches, as outlined in Table 7. Each of the other approaches has some redeeming feature, but
30 One could map the domain to, e.g., [0, 1], but in simple examples we did not see any improvement by doing so.
Also, it might seem confusing that we need to impose an IR cutoff when the background is smooth in the IR. Just as in
AdS, though, the equations of motion are singular for = 0, but the singularity is regular. However, some of the regular
boundary conditions coincide for exactly zero, so to implement them faithfully we need IR > 0.
31 See, e.g., http://www.nr.com.
43
Table 7
Sketch of numerical approaches. Stable means numerical stability, in the empirical sense of Table 6, for the examples we
studied. Multifield means that it generalizes easily to systems with many fields. Avoids Exact means one does not need
exact analytical solution of approximations of the equations of motion in the UV or IR region, or both. No fitting means
one does not have to fit the numerical solutions to prescribed analytical behavior
Method
Stable
Multifield
Avoids Exact
No fitting
Standard shooting
Blind shooting [50, Appendix B]
UV IR Determinant
IR UV Determinant
Midpoint Determinant
/

UV

IR
UV

for computing the KS spectrum, the midpoint method was superior. As can be seen from the
table, the main challenge in using the midpoint method is that one needs to find approximate
asymptotic analytical solution in both the UV and the IR, but this can be done using symbolic
computation.32 We would expect that for backgrounds more complicated than KS, the midpoint
method will remain the most efficient of those in Table 7.
References
[1] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories, Adv. Theor. Math.
Phys. 2 (1998) 505532, hep-th/9803131.
[2] C. Csaki, H. Ooguri, Y. Oz, J. Terning, Glueball mass spectrum from supergravity, JHEP 9901 (1999) 017, hepth/9806021.
[3] R. de Mello Koch, A. Jevicki, M. Mihailescu, J.P. Nunes, Evaluation of glueball masses from supergravity, Phys.
Rev. D 58 (1998) 105009, hep-th/9806125.
[4] J. Polchinski, M.J. Strassler, Hard scattering and gauge/string duality, Phys. Rev. Lett. 88 (2002) 031601, hepth/0109174.
[5] O. Aharony, The non-AdS/non-CFT correspondence, or three different paths to QCD, hep-th/0212193.
[6] A. Zaffaroni, RTN lectures on the non-AdS/non-CFT correspondence, PoS RTN2005 (2005) 005.
[7] M. Krasnitz, A two point function in a cascading N = 1 gauge theory from supergravity, hep-th/0011179.
[8] E. Caceres, R. Hernandez, Glueball masses for the deformed conifold theory, Phys. Lett. B 504 (2001) 6470,
hep-th/0011204.
[9] X. Amador, E. Caceres, Spin two glueball mass and glueball regge trajectory from supergravity, JHEP 0411 (2004)
022, hep-th/0402061.
[10] E. Caceres, C. Nunez, Glueballs of super YangMills from wrapped branes, JHEP 0509 (2005) 027, hepth/0506051.
[11] M. Berg, M. Haack, W. Mck, Bulk dynamics in confining gauge theories, Nucl. Phys. B 736 (2006) 82132,
hep-th/0507285.
[12] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: Duality cascades and chiSB-resolution of
naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[13] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau singularity, Nucl. Phys.
B 536 (1998) 199218, hep-th/9807080.
[14] C.P. Herzog, I.R. Klebanov, P. Ouyang, Remarks on the warped deformed conifold, hep-th/0108101.
[15] M. Strassler, The duality cascade, hep-th/0505153.
[16] J.M. Maldacena, C. Nuez, Towards the large N limit of pure N = 1 super YangMills, Phys. Rev. Lett. 86 (2001)
588591, hep-th/0008001.
[17] A.H. Chamseddine, M.S. Volkov, Non-abelian bps monopoles in N = 4 gauged supergravity, Phys. Rev. Lett. 79
(1997) 33433346, hep-th/9707176.
32 For example, Maple can output optimized Fortran or C code implementing those solutions.
44
[18] A.H. Chamseddine, M.S. Volkov, Non-abelian solitons in N = 4 gauged supergravity and leading order string
theory, Phys. Rev. D 57 (1998) 62426254, hep-th/9711181.
[19] A. Karch, E. Katz, D.T. Son, M.A. Stephanov, Linear confinement and AdS/QCD, Phys. Rev. D 74 (2006) 015005,
hep-ph/0602229.
[20] T. Sakai, J. Sonnenschein, Probing flavored mesons of confining gauge theories by supergravity, JHEP 0309 (2003)
047, hep-th/0305049.
[21] S. Kuperstein, Meson spectroscopy from holomorphic probes on the warped deformed conifold, JHEP 0503 (2005)
014, hep-th/0411097.
[22] D.Z. Freedman, C. Nuez, M. Schnabl, K. Skenderis, Fake supergravity and domain wall stability, Phys. Rev. D 69
(2004) 104027, hep-th/0312055.
[23] K. Skenderis, P.K. Townsend, Hidden supersymmetry of domain walls and cosmologies, Phys. Rev. Lett. 96 (2006)
191301, hep-th/0602260.
[24] R.C. Brower, J. Polchinski, M.J. Strassler, C.-I. Tan, The pomeron and gauge/string duality, hep-th/0603115.
[25] D. Baumann, A. Dymarsky, I.R. Klebanov, J. Maldacena, L. McAllister, A. Murugan, On D3-brane potentials in
compactifications with fluxes and wrapped D-branes, hep-th/0607050.
[26] I.R. Klebanov, A.A. Tseytlin, Gravity duals of supersymmetric SU(N ) SU(N + M) gauge theories, Nucl. Phys.
B 578 (2000) 123138, hep-th/0002159.
[27] T.J. Hollowood, S.P. Kumar, An N = 1 duality cascade from a deformation of N = 4 susy YangMills, JHEP 0412
(2004) 034, hep-th/0407029.
[28] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66 (2002)
106006, hep-th/0105097.
[29] N. Evans, A. Tedder, Perfecting the ultra-violet of holographic descriptions of QCD, hep-ph/0609112.
[30] K. Skenderis, M. Taylor, KaluzaKlein holography, JHEP 0605 (2006) 057, hep-th/0603016.
[31] M. Bianchi, M. Prisco, W. Mck, New results on holographic three-point functions, JHEP 0311 (2003) 052, hepth/0310129.
[32] W. Mck, M. Prisco, Glueball scattering amplitudes from holography, JHEP 0404 (2004) 037, hep-th/0402068.
[33] W. Mck, Progress on holographic three-point functions, Fortschr. Phys. 53 (2005) 948954, hep-th/0412251.
[34] A. Celi, A. Ceresole, G. DallAgata, A. Van Proeyen, M. Zagermann, On the fakeness of fake supergravity, Phys.
Rev. D 71 (2005) 045009, hep-th/0410126.
[35] M. Zagermann, N = 4 fake supergravity, Phys. Rev. D 71 (2005) 125007, hep-th/0412081.
[36] M. Bianchi, D.Z. Freedman, K. Skenderis, Holographic renormalization, Nucl. Phys. B 631 (2002) 159194, hepth/0112119.
[37] D. Martelli, W. Mck, Holographic renormalization and Ward identities with the HamiltonJacobi method, Nucl.
Phys. B 654 (2003) 248276, hep-th/0205061.
[38] I. Papadimitriou, K. Skenderis, AdS/CFT correspondence and geometry, hep-th/0404176.
[39] I. Papadimitriou, K. Skenderis, Correlation functions in holographic RG flows, JHEP 0410 (2004) 075, hepth/0407071.
[40] G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2-sphere, Class. Quantum
Grav. 18 (2001) 13331354, hep-th/0012034.
[41] S.S. Gubser, A.A. Tseytlin, M.S. Volkov, Non-abelian 4-d black holes, wrapped 5-branes, and their dual descriptions, JHEP 0109 (2001) 017, hep-th/0108205.
[42] M. Krasnitz, Correlation functions in supersymmetric gauge theories from supergravity fluctuations, PhD Thesis,
Princeton University, UMI-30-68792, 2003.
[43] H. Firouzjahi, S.H.H. Tye, The shape of gravity in a warped deformed conifold, JHEP 0601 (2006) 136, hepth/0512076.
[44] M. Krasnitz, Correlation functions in a cascading N = 1 gauge theory from supergravity, JHEP 0212 (2002) 048,
hep-th/0209163.
[45] O. Aharony, A. Buchel, A. Yarom, Holographic renormalization of cascading gauge theories, hep-th/0506002.
[46] O. Aharony, A. Buchel, A. Yarom, Short distance properties of cascading gauge theories, JHEP 0611 (2006) 069,
hep-th/0608209.
[47] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, The supergravity dual of N = 1 super YangMills theory, Nucl.
Phys. B 569 (2000) 451469, hep-th/9909047.
[48] M. Bianchi, D.Z. Freedman, K. Skenderis, How to go with an RG flow, JHEP 0108 (2001) 041, hep-th/0105276.
[49] W. Mck, Correlation functions in holographic renormalization group flows, Nucl. Phys. B 620 (2002) 477500,
hep-th/0105270.
[50] M. Berg, H. Samtleben, Holographic correlators in a flow to a fixed point, JHEP 0212 (2002) 070, hep-th/0209191.
The two loop crossed ladder vertex diagram

with two massive exchanges
U. Aglietti a, , R. Bonciani b , L. Grassi c , E. Remiddi c
a Dipartimento di Fisica, Universit di Roma La Sapienza and INFN, Sezione di Roma, I-00185 Rome, Italy
b Departament de Fsica Terica, IFIC, CSIC Universitat de Valncia, E-46071 Valncia, Spain
c Dipartimento di Fisica dellUniversit di Bologna, and INFN, Sezione di Bologna, I-40126 Bologna, Italy
Received 4 June 2007; accepted 23 July 2007

Abstract
We compute the (three) master integrals for the crossed ladder diagram with two exchanged quanta of
equal mass. The differential equations obeyed by the master integrals are used to generate power series
expansions centered around all the singular (plus some regular) points, which are then matched numerically
with high accuracy. The expansions allow a fast and precise numerical calculation of the three master
integrals (better than 15 digits with less than 30 terms in the whole real axis). A conspicuous relation with
the equal-mass sunrise in two dimensions is found. Comparison with a previous large momentum expansion
is made finding complete agreement.
Keywords: Feynman diagrams; Multi-loop calculations
1. Introduction
Electronpositron linear colliders of next generation with a large energy, E = O(1 TeV),
and a large luminosity, L = O(1034 cm2 s1 ), will produce an incredible amount of accurate
data [1]. Theoretical work in the same direction, i.e. pointing to accurate computations of electroweak cross sections and decay widths, is therefore mandatory. That means, in practice, to go
beyond the standard one-loop approximation and address the control of the two-loop quantum
E-mail addresses: ugo.aglietti@roma1.infn.it (U. Aglietti), roberto.bonciani@ific.uv.es (R. Bonciani),

lorenzo.grassi@bo.infn.it (L. Grassi), ettore.remiddi@bo.infn.it (E. Remiddi).
doi:10.1016/j.nuclphysb.2007.07.019
46
U. Aglietti et al. / Nuclear Physics B 789 (2008) 4583
corrections. While in QCD multi-loop computations are often feasible by setting to zero the small
quark masses simplifying then the problem to a massless one, in the electroweak case setting to
zero the masses of the W , Z, Higgs or top quark is not possible. One has to face the general
problem of computing massive multi-scale Feynman diagrams. In many cases, the full computation can be drastically simplified by finding a small expansion parameter , such as the ratio of
particle masses and/or kinematic invariants:
m2
1.
(1)
M2
The heavy algebra entering the higher orders in can be dealt with by using standard algebraic
programs such as Form [2] or Mathematica [3]. In problems involving many different scales or,
equivalently, many different dimensionless ratios, that is probably the only viable analytical way.
Another strategy is that of numerical computation of the integrals from the beginning, such as
for example the numerical computation of loop integrals written in terms of Feynman parameters [4]. This approach is not easy to implement when a subset of particles is massless, because
of troublesome infrared effects. A different strategy assumes the dominance of the logarithmic
terms over constants and power-suppressed terms in the high-energy cross-sections, i.e. of terms
of the form [5]
2 n
s
g
logk 2 (k = 1, 2, . . . , 2n),
(2)
2
16
m
=
where g is the weak coupling and m is the W or Z 0 mass. The logarithmic terms are computed
directly by means of asymptotic expansions [6,7]. This strategy is similar in spirit to the one
used in QCD to construct for example shower Montecarlo programs [8]. The logarithmic terms
are certainly the dominant ones in the formal limit s (i.e. s m), but at a fixed energy nonlogarithmic terms may in general have a significant numerical effect. In cases in which one or
(at most) two dimensionless quantities of order one are involved, an exact analytic computation
may be useful. A model computation for higher-order electroweak effects is the two-loop form
factor in the degenerate mass limit,
mW = mZ = mH = m.
(3)
In [9,10] two of us have computed the master integrals of the two-loop electroweak form factor for all diagrams with the exception of the crossed-ladder with the exchange of two equal
mass quantasuch as a pair of W s and/or Zs (see Fig. 1). These master integrals are naturally expressed in the basis of the harmonic polylogarithms (HPLs), transcendental functions
verifying algebra properties which largely simplify the transformations [11]. Some master integrals involving two or three massive propagators have required a generalization of the harmonic
polylogarithms basis, involving basic functions with square rootsthe so-called generalized harmonic polylogarithms (GHPLs) [10].
In this paper we compute the last missing piece in order to obtain the complete two-loop
electro-weak form factor: the crossed ladder diagram with two equal-mass quanta exchanged. By
standard reduction techniques, we find that this topology involves three master integrals. At variance with respect to the planar topologies and the non-planar topology with at most one massive
exchange, it is not possible to represent any of these master integrals in terms of harmonic polylogarithms or generalized harmonic polylogarithms: elliptic integrals are involved, i.e. integrals
of square roots of polynomials of degree three or four [12]. For that reason, we use a different
method, already applied in [13,14] to evaluate the equal-mass sunrise diagram. The differential
47
Fig. 1. Feynman diagram for the annihilation of a pair of massless fermions with the exchange of two massive quanta
with equal mass m. The thin lines represent the massless fermions, while the thick lines represent the massive quanta.
The outgoing dashed line represents the probe, denoted generically as Z .
equations obeyed by the master integrals are used to generate power-series expansions centered
around all the singular points of the differential equations themselves. It may also be convenient
to generate auxiliary expansions around regular points, which are in principle completely arbitrary. The series expansions centered around two different points are then matched numerically
in a point belonging to the intersection of the respective domains of convergence. In general, we
are not able to find a closed analytic expression for the master integrals, but only truncated series
representations. The latter allow, however, a fast and accurate numerical evaluation of the master
integrals themselves (better than 15 digits with less than 30 terms, typically, for the accelerated
versions of the expansions). The logarithmic terms in the large momentum expansion of one of
the master integrals (the basic scalar amplitude) have been obtained in [15]. Our expansion is in
agreement with these results; we are also able to compute the power-suppressed corrections.
2. Power series solution of the differential equations on the MIs
In this section we outline the computation of the master integrals of the two-equal-massexchanged crossed-ladder diagram, which is described extensively in the next sections. As
anticipated in the introduction, we compute the master integrals with a general semi-analytical
method [12], which can be applied to arbitrarily complicated cases. The first steps (Section 4) are
common to the method used to compute the previous electroweak form factor diagrams [9,10].
By using standard reduction procedures, the problem is shifted to that of computing three independent scalar amplitudes, called master integrals (MIs), which obey a system of coupled
2 is the
first-order differential equations in the evolution variable x s/m2 , where s = Ecm
Mandelstam variable (Section 5). These equations are linear and not homogeneous. By a suitable
choice of the master integral basis, we succeeded in triangularizing the system to a system of two
coupled equations (let us say the first two equations on F1 and F2 ) plus a decoupled equation
from the previous ones (let us say the third one on F3 ):
dF1
= A11 (x)F1 (x) + A12 (x)F2 (x) + 1 (x);
dx
dF2
= A21 (x)F1 (x) + A22 (x)F2 (x) + 2 (x);
dx
(4)
(5)
48
dF3
(6)
= A31 (x)F1 (x) + A32 (x)F2 (x) + A33 (x)F3 (x) + 3 (x),
dx
where the Aij (x)s, the coefficients of the associated homogeneous system, are rational fractions
in x, while the k (x)s are known functions (logarithms, polylogarithms, etc.). We then derive
from the system of the two coupled first-order equations a single second-order equation in one
of the two master integrals involved, let us say F1 (Section 6):
d 2 F1
dF1
(7)
+ A(x)
+ B(x)F1 (x) = (x).
dx
dx 2
The resulting equation belongs to the Fuchs class, i.e. it has regular singular points only, including also the point at infinity1 [12]. There are four such singularities, all located on the real axis,
at:
x = 0;
x = 8;
x = 1;
x = .
(8)
The differential equation (7) is used to generate power series expansions centered around all these
singular points (Section 7). Auxiliary expansions around (arbitrary) regular points may also be
of utility. Each series is determined by the differential equation up to two arbitrary coefficients,
which must be fixed by some initial or boundary condition. A qualitative knowledge of the behavior of the master integral F1 (x) around x = 0 is sufficient to fix exactly (in fact analytically)
the two arbitrary coefficients of the expansion around this point. The coefficients of the series
centered around x = 0 (small-momentum expansion) are therefore completely determined in analytical way. The arbitrary coefficients occurring in the series centered in the remaining points
are found by imposing matching conditions in suitable points belonging to the intersections of
the respective domains of convergence. Assume that one knows the expansions of F1 (x) around
two different points x1 and x2 ,
F1 (x) =
F1 (x) =

n=0
an(1) (x x1 )n
for |x x1 | < R1 ,
(9)
an(2) (x x2 )n
for |x x2 | < R2 ,
(10)
n=0
(1)
with |x1 x2 | < R1 + R2 , so that the two circles overlap, and that the coefficients an of the
first series are completely known, while those of the second series are determined as (linear)
functions of the first two by the underlying differential equation:
an(2) = bn a0(2) + cn a1(2) + dn .
(11)
By taking a point x in the intersection domain, the matching conditions give a linear system on
(2)
(2)
the unknowns a0 and a1 .
a0(2)

n=0
bn (x x2 )n + a1(2)

n=0
cn (x x2 )n =

n=0
an(1) (x x1 )n
dn (x x2 )n ;
(12)
n=0
1 In general, by a singularity of the differential equation we mean a singularity in the coefficients A(x) or B(x).
A point x0 is a regular singular point of a differential equation if its solutions can be written as singular factor (x x0 )
or (x x0 ) log(x x0 ) multiplied by a convergent series in a neighborhood of x0 .
49
(2)
a0
nbn (x x2 )
n=1
n1
(2)
+ a1
ncn (x x2 )n1
n=1
nan(1) (x x1 )n1
n=1
(2)
ndn (x x2 )n1 .
(13)
n=1
(2)
(2)
Once a0 and a1 are known, one determines the higher-order coefficients an for n > 1. Instead
of matching the derivatives of the series, one can also take a second matching point x .
Once the first master integral F1 has been determined, one plugs its expression into the
first-order equation (4) of the 2 2 system involving dF1 /dx and determines algebraically F2
(Section 8). One finally inserts the expression of the first two MIs F1 and F2 into the third differential equation (6) and determines F3 by quadrature. Let us remark that one can study all the
analytical properties of the MIs (analytic continuation, behavior close to thresholds, asymptotic
properties, etc.) by looking at their power-series expansions. The latter also offer a suitable and
powerful way for the precise and fast numerical evaluation of the MIs, for example with a Fortran
routine.2
In Section 9 we discuss the relation between the differential equation on the master integral F1
for the crossed ladder and that one for the sunrise diagram with three equal masses in two space
time dimensions [13]. The physical origin of this connection is, at present, unclear. Finally, in
Section 10 we draw our conclusions.
3. Threshold structure
The crossed ladder diagram with two equal-mass quanta exchanged has a rather simple threshold structure: there is a 2-particle cut on the internal massless lines for
s0
(14)
and two 3-particle cuts on the internal massless lines and one of the massive lines for
s m2 .
(15)
There is also a pseudo-threshold for s

The complexity of the differential equations and
of the structure of the master integrals cannot be guessed by looking at the threshold structure of
the diagram. That is to be contrasted to the case of the equal-mass sunrise, where the threshold
at s = 9m2 as well as the pseudo-threshold in s = m2 are identifiable as the basic sources of
complexity.
4m2 .
4. Reduction to master integrals

By standard decomposition into invariant form factors and rotation of the scalar products,
one can show that the computation of the two equal-mass crossed ladder diagram (see Fig. 1) is
equivalent to the computation of the following 7 independent scalar amplitudes:

Sr
D
D
F (n1 , n2 , n3 , n4 , n5 , n6 , r) =
(16)
n1 n2 n3 n4 n5 n6 D k1 D k2 ,
P1 P2 P3 P4 P5 P6
2 The execution time for the numerical evaluation of the resulting expansions is expected to be essentially the same as
the time required for the expansions discussed in [14], that is less than 1 microsecond for the three integrals on a 3 GHz
CPU.
50
where D is the spacetime dimension, the scalar product is defined as

a b a
b
a0 b0 ,
(17)
the loop measure is

DD k
1
dDk
,
(3 D/2) 4 D/2
(18)
with (z) the Euler Gamma function. We consider a routing of the loop momenta k1 and k2
which results in the following denominators:
P1 = k12 + m2 ,
P2 = k22
(19)
+m ,
2
(20)
P3 = (p1 k1 ) ,
2
(21)
P4 = (p2 k2 ) ,
2
(22)
P5 = (p1 k1 + k2 ) ,
(23)
P6 = (p2 + k1 k2 ) ,
(24)
2
2
and the following irreducible numerator (scalar product):

S = p2 k1 .
(25)
The indexes of the denominators are assumed to be all positive,3 ni > 0 while the index of the
scalar product can be positive or zero, r 0.
The above independent scalar amplitudes, considered as functions of the integer indexes ni s
and r, are related to each other by integral identities obtained in general by means of integration by parts over the loop momenta, invariance under Lorentz transformations and symmetry
relations coming from the particular mass distribution of the diagram under consideration [16].
These identities can be used to express a given amplitude F as a linear combination of a set of
reference amplitudes, with fixed indexes, called master integrals (MIs) Fi :
F (n1 , n2 , n3 , n4 , n5 , n6 , r) =
N
ci (n1 , n2 , n3 , n4 , n5 , n6 , r)Fi ,
(26)
i=1
where the ci s are known coefficients (rational fractions in x and D). The above reduction involves the solution of the integration-by-parts identities in some recursive way over the indexes.
Nowadays the most common algorithm is the Laporta algorithm [17], which we have used for
the crossed ladder topology. That way we have been able to reduce the independent amplitudes
to a linear combination of three master integrals:
F (n1 , n2 , n3 , n4 , n5 , n6 , s) =
3
ci (n1 , n2 , n3 , n4 , n5 , n6 , s)Fi + ,
(27)
i=1
where the dots denote contributions from master integrals with less than six denominators, i.e.
from sub-topologies, which are known [9,10,18]. Let us choose the following basis of master
3 If n 0 for some i we have a sub-topology in which line i is shrink to a point.
i
51
integrals:
1
D D k1 D D k2 ;
P1 P2 P3 P4 P5 P6

1
F2 (x; D) =
D D k1 D D k2 ;
2
P1 P2 P3 P4 P5 P6

S
D D k1 D D k2 .
F3 (x; D) =
P1 P2 P3 P4 P5 P6
F1 (x; D) =
(28)
(29)
(30)
The first MI is the basic scalar amplitude, with linear denominators and no scalar products; the
second MI contains the first denominator squared, while the third MI involves an irreducible
numerator. By direct inspection, one finds that the above MIs do not contain any pole in D 4
because they are ultraviolet finite and, since the radiated quanta are massive, there is an effective
infrared cut-off m which renders the MIs infrared finite as well. Let us therefore set D = 4
from now on and omit this argument.
5. The system of differential equations
Once the reduction to master integrals has been achieved, the following step is the actual
computation of the master integrals themselves. We use the differential equation method [19],
which involves taking a derivative with respect to
x=
s
,
m2
(31)
at fixed m2 , of the master integral Fi :

x
Fi
Fi
Fi
= p1 = p2 .
x
p1
p2
(32)
q = p1 + p2 is the momentum of the probe and q 2 = s, s being the squared c.m. energy. By
taking the derivative inside the integral in Eq. (32), various amplitudes are generated, which are
reduced again to master integrals as discussed in the previous section. We then obtain in general
a system of linear differential equations with variable coefficients, of the form:

d
fij (x)Fj (x) + Ni (x)
Fi (x) =
dx
3
(i = 1, 2, 3).
(33)
j =1
By changing the basis for the master integrals, the functions fij (x) and Ni (x) are transformed
into new functions. A general basis for the MIs involves a system of three coupled differential
equations, which are equivalent to a single, third-order differential equation, whose solutions, as
well known, are rather difficult to find. With the basis given in the previous section, the system
comes out to be triangular with a 2 2 block. The coefficients of the associated homogeneous
system read, in this basis:
2
2
f11 (x) = ; f12 (x) = ; f13 (x) = 0;
x
x
1
1
1
1
1
+
; f22 (x) =
;
f21 (x) = +
2x 3(x + 1) 6(x 8)
x +1 x 8
(34)
f23 (x) = 0;
(35)
52
1
f31 (x) = ;
6x

1
1
f32 (x) =
1+
;
3
x
1
f33 (x) = ;
x
(36)
while the known terms, related to the sub-topologies, are:

N1 (x) = 0;
N2 (x) =
1 1
9
1
H (0, 0; x) H (0, 1; x) +
H (r, 0; x) +
16x 2
4
4
x(4 x)

1
1
5
1
H (r, 0; x) +
H (0, 0; x) H (0, 1; x) +
4(x + 1) 9
2
12 x(4 x)

1
1
1
1
H (r, 0; x) +
H (0, 0; x) H (0, 1; x)
8(x 8) 36
8
3 x(4 x)
N3 (x) =

1
2
1
H
(1,
0,
0;
x)
+
H
(r,
r,
0;
x)
+
H
(1;
x)
.
4
48x 2 2
(37)

2
2
18

2
;
72
(38)
(39)
The known terms are written in terms of generalized harmonic polylogarithms [10]. Since
f13 (x) = 0 and f23 (x) = 0, F3 decouples from the system of the first two MIs F1 and F2 .
It is therefore natural to compute F1 and F2 first. Once the latter are known, the third master integral F3 can be computed by quadraturei.e. by integration of the first-order differential
equation (33) with i = 3.
5.1. Master integrals close to zero external momentum
Since p12 = p22 = 0,
q 2 = 2p1 p2
(40)
and the limit

x 0
(41)
is equivalent to the limit p1 p2 0. However, limit (41) also implies the stronger limits p1 0
and p2 0, which reduce the vertex topology to a vacuum one. That can be seen for example by
introducing Feynman parameters x1 , x2 , . . . , x5 for the vertex amplitude and integrating over the
loop momenta k1 and k2 . After that, the integrand can depend on relativistic invariants only, i.e.
2
just on q . One then performs analytic continuation to Euclidean space, where the limit q 2 0
implies q 0, i.e. p2 = p1 . The diagram can then depend only on p12 = 0 and therefore
must be equal to that one computed for p1 = 0, i.e. to the corresponding vacuum amplitude. In
particular, the MIs reduce to the following vacuum amplitudes:

1
D 4 k1 D 4 k2 ;
F1 (x = 0) =
(42)
2
2
2
2
(k1 + m )(k2 + m )k12 k22 [(k1 k2 )2 ]2

1
D 4 k1 D 4 k2 ;
F2 (x = 0) =
(43)
2
2
2
2
(k1 + m ) (k2 + m2 )k12 k22 [(k1 k2 )2 ]2
F3 (x = 0) = 0.
(44)
53
The above integrals can be computed exactly by using the identities

1
1
1
1
.
=
2
2
2
2
2
2
2
ki ki + m
m ki
m (ki + m2 )
(45)
For our purposes, that is actually not necessary. By shrinking the massive lines in the infrared
regions k12 , k22 m2 , we obtain:
1
F1 (x = 0) 4
m
1
F2 (x = 0) 6
m
UV
UV
1
D 4 k1 D 4 k2 ;
k12 k22 [(k1 k2 )2 ]2
(46)
1
D 4 k1 D 4 k2 ,
k2 )2 ]2
(47)
k12 k22 [(k1
where an ultraviolet cutoff UV m is assumed. The integrands above are invariant under the
limit 0, where the parameter is introduced through the rescaling
k1 k1 ,
k2
(48)
k2 .
(49)
divergence.4
All that implies an over-all logarithmic soft

There is also a logarithmic collinear
singularity for k1 k2 related to the denominator [(k1 k2 )2 ]2 . As a consequence5 :

F1,2 (x) log2 x,
F3 (x) 0 for x 0.
(50)
That implies in particular that no pole terms 1/x can appear. As we are going to show explicitly, this qualitative information is sufficient for the solution of the system at small external
momenta. Finally, let us note that we can set m = 1 from now on without loosing any information
(m is an over-all scale).
6. Second-order differential equation for F1 (x)
It is convenient to transform the system of differential equations for F1 and F2 into a single
second-order equation for F1 only. The procedure is standard: one takes an additional derivative
with respect to x on both sides of Eq. (33) with i = 1; dF1 /dx and dF2 /dx are replaced by the
r.h.s. of Eqs. (33) with i = 1 and i = 2, respectively; finally F2 is replaced by its expression in
terms of dF1 /dx and F1 , coming from the first of Eqs. (33). We obtain:
dF1
d 2 F1
(x) + A(x)
(x) + B(x)F1 (x) + C(x) = 0,
dx
dx 2
where:
1
1
3
+
;
A(x) = +
x x +1 x 8
9
1
4
5
B(x) = 2 +
+
;
8x 3(x + 1) 24(x 8)
x
(51)
(52)
(53)
4 There is not any soft sub-divergence, as can be seen by power-counting under the rescaling k k , k k
1
1
2
2
and k1 k1 , k2 k2 .
5 If one sets x = 0 from the very beginning, 1/(D 4) poles of infrared nature do appear in F and F .
1
2
54

1
7
1
1
C(x) =
H (0, 0; x)
16x 2 128x 18(x + 1) 1152(x 8)

63
1
1
9
H (0, 1; x)
+
2
256x 4(x + 1) 256(x 8)
32x

7
5
1
1
1
H (r, 0; x)
+
+
2
32x 24(x + 1) 96(x 8)
8x
x(4 x)

7
1
1
2 1
4 8x 2 64x 9(x + 1) 576(x 8)

Let us make a few remarks:
(54)
1. the coefficient A(x) of F1 (x) contains simple poles, while the coefficient B(x) of F1 (x) contains at most double poles. That is the necessary and sufficient condition on the differential
equation to have regular singular points only [12];
2. the denominators entering the coefficients are:
1
1
1
1
(55)
;
;
;
.
x
x +1
4x
x 8
The last denominator was not expected a priori, as it does not correspond to any threshold
or pseudo-threshold of the diagram. The denominator 1/(4 x) only appears in the known
function C(x) and it is related to the sub-topologies with the pseudo-threshold in s = 4m2 ;
3. the known term C(x) contains GHPLs (see Appendix A) of weight at most two;
4. the GHPL H (r, 0; x),
containing a single square-root basic function, has a coefficient involving a square root 1/ x(4 x) to ensure reality of the solution across the pseudo-threshold
located at x = 4.
7. Power-series solution of the second-order differential equation for F1 (x)
The second-order differential equation (51) for F1 is by far too complicated to be solved in a
closed analytical form. We therefore look for solutions in the form of power series expansions
centered around some points on the real axis. Let us recall that it is in any case necessary to
consider the expansions around all the singular points, as the latter are by definition outside
the convergence region of the ordinary expansions at regular points. Thanks to the differential
equation satisfied by the considered functions, expansions around singular or regular points can
be performed almost in the same way.
Since the coefficients A and B diverge for x = 0, 1, 8, we conclude that these points are
singular points for the differential equation. Also x = is a singular point for the differential
equation, as can be seen by changing variable to y = 1/x and studying the limit y 0.
7.1. Expansion around x = 0small momentum expansion
In this section we consider the expansion of the master integral F1 (x) around x = 0. Since the
nearest singularity to the origin is located in x = 1, we expect a radius of convergence
R0 = 1,
(56)
i.e. convergence in the complex x-plane for

|x| < 1.
(57)
55
For real x, that means:

1 < x < 1.
(58)
Because of linearity, the general solution of the complete inhomogeneous equation can be written as the superposition of the general solution of the associated homogeneous equationi.e.
Eq. (51) with C(x) = 0,
(0)
(0)
dF
d 2 F1
(x) + A(x) 1 (x) + B(x)F1(0) (x) = 0,
dx
dx 2
plus a particular solution of the complete equation:
(0)
F1 (x) = F1 (x) + F1 (x).
(59)
(60)
We will use systematically Eq. (60) in the following expansions.

7.1.1. Homogeneous equation
Let us first consider the associated homogeneous equation. Since zero is a singular regular
point, we can look for solutions having the form of a singular function in x = 0, S(x), multiplied
by a regular (i.e. convergent) power-series6 :
(0)
F1 (x) = S(x)
An x n ,
(61)
n=0
where An are numerical coefficients determined from the differential equation itself and from
some initial or boundary conditions. The function S(x), giving the leading singularity for x 0,
is assumed to be of power-like form
S(x) = x ,
(62)
and solves the limit of the homogeneous equation for x 0:

3
1
S (x) + S (x) + 2 S(x) = 0.
x
x
(63)
By inserting a solution of the form (62), we obtain the indicial equation ( + 1)2 = 0 with a
double zero in = 1, implying two independent pre-factors of the form:
1 log x
,
.
(64)
x
x
The most general solution of the homogeneous differential equation is therefore of the form:
S(x) =
(0)
F1 (x) =

n=1
an x + log x
n
bn x n ,
(65)
n=1
where we have absorbed a 1/x factor inside the series by defining:

an An+1 .
(66)
6 This case is to be contrasted to that of an irregular singular point, in which the singularity of the differential equation
is so strong that no factorization of the singularity of the form (61) is possible. In the latter case, one typically obtains
series with an infinite number of negative powers or asymptotic (divergent) expansions.
56
By expanding the differential equation (59) around x = 0 and substituting the series representation in Eq. (65), we recursively obtain all the desired coefficients. The first few are:
1
3
a0 = a1 b1 ;
4
8
5
33
a1 = a1 +
b1 ;
32
128
7
25
a2 = a1
b1 ;
64
128
173
2561
a3 =
a1 +
b1 ;
2048
16384
Let us make a few remarks:
1
b0 = b1 ;
4
5
b1 = b1 ;
32
7
b2 = b1 ;
64
173
b3 =
b1 .
2048
(67)
(68)
(69)
(70)
Setting to zero for example a1 and b1 , all the higher-order coefficient vanish and we obtain
the trivial solution: that is the homogeneity property.
There is a triangular structure: all the coefficients bi are proportional to the lowest-order one
b1 , implying that by setting b1 = 0 we obtain a solution without the series with the logarithmic pre-factor and with the simple pole in x = 0 only. The coefficients ai depend instead
on both a1 and b1 , implying that by setting a1 = 0 one obtain a solution containing both
series.
A first independent solution can be obtained by taking for example a1 = 1 and b1 = 0,
resulting in a function without logarithmic terms. A second independent solution can be
obtained by taking a1 = 0 and b1 = 1, resulting in a function having both the power and
the logarithmic terms. With a pictorial language, we may say that the logarithmic series
feeds the standard one, while the vice versa is not true.
The singularity of the differential equation at x = 0 produces a logarithmic branch point in
the solution at x = 0 whenever b1 = 0.
7.1.2. Complete equation
Let us now consider the complete equation (51). We have to expand around x = 0 also the
inhomogeneous term C(x), which is known (see Appendix B):

1
n
n
2
n
C(x) = 2
(71)
kn x + log x
qn x + log x
rn x .
x
n=0
n=0
n=0
The known term has a double pole in x = 0, multiplied also by a single or a double logarithm
of x. The expected radius of convergence of the multiplying series is one:
R = 1.
(72)
The explicit expressions of the lowest-order coefficients read:

1 2
;
8 32
23
7 2
k1 =
+
;
288 256
3931
57 2
k2 =
;
28800 2048
k0 =
1
;
16
19
q1 =
;
192
671
q2 =
;
7680
q0 =
1
;
32
7
r1 =
;
256
57
r2 =
;
2048
r0 =
(73)
(74)
(75)
1789247
38791
455 2
q3 =
+
;
;
11289600 16384
430080
Let us look for a particular solution of (51) of the form:
k3 =
F1 (x) =

n=1
pn x n + log x

n=1
qn x n + log2 x
r3 =
cn x n .
455
.
16384
57
(76)
(77)
n=0
By inserting the above form of the solution into the differential equation expanded around x = 0
with the known term C(x) given by the series expansion in Eq. (71), we obtain for the first few
coefficients:
1
3
2
1
p0 = p1 q1 +
(78)
;
4
8
32 16
1
1
q0 = q1 ;
(79)
4
16
1
c0 = ;
(80)
32
5
33
2
5
q1
+
;
p1 = p1 +
(81)
32
128
64 576
5
1
q1 = q1 + ;
(82)
32
96
1
c1 = ;
(83)
64
7
25
13 2
1093
p2 = p1
(84)
q1 +

;
64
128
1152
518400
7
121
q2 = q1
(85)
;
64
17280
13
c2 =
(86)
;
1152
173
2561
5 2
649
p1 +
q1
+
;
p3 =
(87)
2048
16384
576 635040
173
95
q3 =
(88)
q1 +
;
2048
24192
5
c3 =
(89)
;
576
563
42631
407 2
3217
p4 =
(90)
p1
q1 +

;
8192
327680
57600
6720000
563
1141
q4 =
(91)
q1
;
8192
432000
407
c4 =
(92)
.
57600
The coefficients ci of the double-logarithmic terms are completely determined, while the remaining ones pi and qi are fixed once two arbitrary coefficients, such as for example p1 and q1 ,
have been fixed. That is exactly the same arbitrariness that we have already found for the associated homogeneous equation. Since the general solution of the latter has already been found, we
need only to find a particular, i.e. a single solution of the complete equation. Let us choose for
58
example:
p1 = 0 and q1 = 0.
(93)
With the following arbitrary choice, the particular solution does not contain any pole term and
its numerical coefficients are completely fixed:
1
1
2
+
;
q0 = ;
16 32
16
5
1
2
;
q1 = ;
p1 =
576 64
96
1093
121
13 2
+
;
q2 =
;
p2 =
518400 1152
17280
649
95
5 2
,
q3 =
;
p3 =
635040 576
24192
3217
1141
407 2
+
;
q4 =
,
p4 =
6720000 57600
432000
The general solution of the complete equation is therefore:
p0 =
F1 (x) =
an x n + log x
n=1

n=1
bn x n +
1
;
32
1
c1 = ,
64
13
c2 =
;
1152
5
c3 =
,
576
407
c4 =
.
57600
c0 =
pn x n + log x
n=0

n=0
qn x n + log2 x
(94)
(95)
(96)
(97)
(98)
cn x n .
n=0
(99)
To uniquely determine the solution, one has to impose some boundary or initial conditions. As
already discussed in the previous section, F1 (x) can have at most a logarithmic singularity for
x 0, implying that the coefficients of the power singularities must vanish:
a1 = 0,
b1 = 0,
(100)
and the solution of the homogeneous equation to be selected, reduces to the trivial one. The
particular solution F1 of the complete equation that we have chosen is therefore the expansion of
the master integral F1 as defined by Eq. (28):
F1 (x) =
pn x n + log x
n=0

n=0
qn x n + log2 x
cn x n ,
(101)
n=0
where the coefficients are given in Eqs. (94)(98).

Let us stress that we have been able to obtain the complete analytical expression of the
coefficients of this small-momentum expansion because of the knowledge of the MI at small
momentum transferred. As we are going to show in the next sections, the absence of a similar
knowledge in other expansion points will limit us to a numerical estimate of the corresponding
coefficients.
As expected from the threshold structure, the solution given in Eq. (101) is real for x > 0
(space-like region) and has a non-vanishing imaginary part for x < 0 (time-like region). The
latter can be determined by simply giving the prescription for the log x factor. Since s s + i
with = +0, Eq. (31) gives for x < 0:
x |x| i.
(102)
59
Consequently, we have:
log x log |x| i;
log2 x log2 |x| 2 2i log |x|.
(103)
Therefore F1 (x) becomes complex for x < 0 with:

Re F1 (x) =

n=0
pn x n + log |x|

Im F1 (x) =
qn x n + log2 |x| 2
cn x n ;
n=0
qn x + 2 log |x|
n
n=0

cn x
(104)
n=0
(105)
n=0
In the following sections, we will move on the real axis constructing the solution from x = 0
to x = ; then with an analytic continuation we will switch to x = and finally move back
to x = 0 through negative values of x.
7.1.3. Improved expansionBernoulli variable
A series for F1 with better convergence properties than the previous one in x can be constructed by expanding in the Bernoulli variable [20]

4095 4
32769 5
9
63 2 171 3
1+x
= x
x +
x
x +
x + O x6
t log 8
8x
8
128
512
16384
163840

2
3
4
= 1.125x 0.492187x + 0.333984x 0.249938x + 0.200006x 5 + O x 6 .
(106)
Eq. (106) realizes a one-to-one map between x (1, 8) and t (, ). Furthermore,
t =0
x = 0,
(107)
while
t for x 1+ ,
t + for x +8 .
(108)
t diverges logarithmically when x approaches the singularities of the differential equation closest
to the origin. Since this variable follows the singularities of the differential equation, we expect
a faster convergence with the order of truncation in t rather than in x. The inverse of Eq. (106)
reads:
x=
8(et 1) n
cn t .
=
et + 8
(109)
n=1
The first few terms are:

8
44 3
35 4
1733 5
28
x = t + t2 +
t
t
t + O t6
9
81
729
6561
295245
2
= 0.888888t + 0.345679t + 0.0603566t 3 0.00533455t 4

0.00586970t 5 + O t 6 .
(110)
60
The radius of convergence of the above series is determined by looking at the singularities of
x = x(t) in the complex t -plane, located in
tk = 3 log 2 + i(2k + 1),
(111)
where k is an integer. The closest singularities to the origin, for k = 0, 1, give

r0 =
2 + 9 log2 2 3.76745.
(112)
We now substitute the r.h.s. of Eq. (109) in the series expansion for F1 (x) obtained previously
and finally expand in powers of t , to have7 :

n t n + log x
n t n + log2 x
n t n ,
F1 (t) F1 x(t) =
n=0
n=0
(113)
n=0
where the coefficients n , n and n are determined from an , bn and cn , respectively. The first
few orders read:
0 =
1
2
+
;
16 32
0 =
1 =
5
2
;
648 72
1 =
2 =
3503
41
+
2;
2624400 11664
2 =
1
;
16
1
;
108
169
;
87480
0 =
1
;
32
1
;
72
(115)
41
.
11664
(116)
1 =
2 =
(114)
We have computed F1 (t) as a function of t (Eq. (113)) by substituting the series (109) into the
series (101). The problem now is that of computing the radius of convergence 0 of the series
expansion for F1 (t). The internal series for x = x(t) converges inside the circle in the t -plane
of radius r0 given in Eq. (112); the external series for F (x) converges in the unitary circle
|x| < 1 in the x-plane (Eq. (57)). As well known, power series always converge inside circles.
The point is that a circle in the t -plane in general is not mapped by the function x = x(t) into
a circle in the x-plane. Our problem has the following geometrical formulation: one has to find
the largest circle in the t -plane satisfying |t| < r0 , which is mapped inside the circle of unitary
radius in x-plane by the transformation x = x(t) [21]. On the unitary circle x = exp(i) and one
has to look for a minimum over of

1 + exp(i)
16
|t| = log 8
log
< r0 ,
8 exp(i)
7
(117)
7 Let us note that we did not express the logarithmic pre-factors log x and log2 x as series in t , because there was no
practical advantage for doing that.
61
where the minimum is obtained for = 0. We then find8 :

16
0.826679.
7
0 = log
(120)
In general, for a series expansion centered in x0 with nearest singularities a and b satisfying
(121)
a < x0 < b,
the Bernoulli variable t is defined as:

b x0 x a
t = log
.
x0 a b x
(122)
Note that
t =0
x = x0 ,
(123)
while
t for x a + ;
t + for x b .
(124)
The following particular case are relevant in the following:

x0 :

x a
t = log
;
x b
(125)
b :

x a
;
t = log
x0 a
(126)
a :

x b
.
t = log
x0 b
(127)
8 We have shown that the convergence radius of the t -series of F is not smaller than , but it can actually be larger,
1
0
depending on possible elimination of singularities. That can be illustrated with the following (rather trivial) example. Let
us consider the expansion around x = 0 of a differential equation having a solution of the form
(x) = log
8(1 + x)
.
8x
(118)
If we go to the Bernoulli variable t defined in Eq. (106), we obtain

= x(t) = t,
(t)
which can be analytically continued to all the t -plane, implying an infinite radius of convergence.
(119)
62
7.2. Expansion around x = 8

In this section we consider the expansion of F1 (x) around the (space-like) point x = 8, which
is a regular singular point of the differential equation not corresponding to any threshold or
pseudo-threshold of the diagram. The singular point of F1 (x) closest to x = 8 is located in x = 0.
We then expect the expansion around x = 8 to have a radius of convergence
R8 = 8,
(128)
so that the series converges in the complex x-plane for

|x 8| < 8.
(129)
For real x, that means:

0 < x < 16.
(130)

By solving the indicial equation as in the previous section, we obtain a double zero in zero,
so that the homogeneous equation has a solution of the form:
F1(0) (x) =
an (x 8)n + log(x 8)
n=0
bn (x 8)n .
(131)
n=0
The coefficients are, of course, different from those ones of the previous sectionwe use the
same notation to avoid introducing too many symbols. The first few coefficients read:
5
5
a0 b0 ;
24
72
59
187
a0 +
b0 ;
a2 =
1728
10368
635
7561
a0
b0 ;
a3 =
124416
2239488
2171
59447
a4 =
a0 +
b0 ;
2985984
107495424
a1 =
5
b0 ;
24
59
b0 ;
b2 =
1728
635
b0 ;
b3 =
124416
2171
b4 =
b0 .
2985984
b1 =
(132)
(133)
(134)
(135)
We have again a triangular structure of the coefficients, as in the previous case.

The expansion of the known term around x = 8 is of the following form:
C(x) =
qn (x 8)n ,
(136)
n=1
where the first three coefficients qn are:

q1 =
1
1
1
1
1
2 +
M0 2
M1 +
M2 2 log 2 +
log2 2;
2304
768
256
192
256
(137)
13
29
13
29
1
2
M0 2 +
M1
M2 2 log 2 +
log 2
82944
55296
9216
13824
768
1
13
log2 2
log 3;
9216
1024
451
4637
451
4637
2 +
M0 2
M1 +
M2 2 log 2
q1 =
11943936
31850496
1327104
7962624
61
551
451
2
log 2 +
log 2 +
log 3.
884736
1327104
147456
We have defined the following transcendental constants:
63
q0 =
1
M0
0
(138)
(139)
log(1 + y)
dy
y(1 + y)

2
+ 4 log 2 log( 2 1) + 2 log2 ( 2 1) + 4 Li2 i( 2 1)
6

+ 4 Li2 i( 2 1)
0.425435,

1
2 9
+ log2 2 + Li2
3.68568;
M1 =
6
2
8
M2 = log(1 + 2) 0.881374.
(140)
(141)
(142)
The general solution of the inhomogeneous equation reads:

F1 (x) =
rn (x 8) + log(x 8)
n
n=0
pn (x 8)n ,
(143)
n=0
with the first few terms given by:
1
1
1
1
1
2
M0 2 +
M1
M2 2 log 2
log2 2
2304
768
256
192
256
5
5
r0 p0 ;
(144)
24
72
5
p1 = p 0 ;
(145)
24
19
79
19
79
1
2 +
M0 2
M1 +
M2 2 log 2
log 2
r2 =
165888
221184
18432
55296
3072
59
19
1
187
log2 2 +
log 3 +
r0 +
p0 ;
+
(146)
18432
4096
1728
10368
59
p0 .
p2 =
(147)
1728
A particular solution can be obtained by setting p0 = 0, which makes all the coefficients pn
vanishing: that way the series multiplied by the logarithm disappears from F1 . One can also set
r0 = 0. The first three coefficients are then given by:
r1 =
r0 = 0;
(148)
p0 = 0;
(149)
64
1
1
1
1
1
2
M0 2 +
M1
M2 2 log 2
log2 2;
2304
768
256
192
256
p1 = 0;
19
79
19
79
1
r2 =
2 +
M0 2
M1 +
M2 2 log 2
log 2
165888
221184
18432
55296
3072
19
1
+
log2 2 +
log 3;
18432
4096
p2 = 0.
r1 =
(150)
(151)
(152)
(153)
The general solution of the differential equation is given by

F1 (x) = F1 (x) + F1 (x),
(0)
(154)
(0)
and depends on the arbitrary constants a0 and b0 entering F1 (x). Since we know from general
arguments that x = 8 is a regular point for the solution, we can impose the logarithmic series be
absent by requiring
b0 = 0.
(155)
This condition gives:

F1 (x) =
sn (x 8)n ,
(156)
n=0
where:
s0 = a 0 ;
(157)
5
s1 = a0 + r1 ;
(158)
24
59
a0 + r2 ,
s2 =
(159)
1728
etc. Then, the expansion of F1 (x) around x = 8 given in Eq. (156) does not determine F1 (x)
uniquely, because it still contains the free parameter a0 . By using the matching procedure described in Section 7.2.4, we obtain the following numerical estimate for this coefficient:
a0 = 0.0321062814000779405116.
(160)
7.2.3. Improved expansion

In order to improve the convergence of the series so far obtained, we move from the series in
x to the one in the Bernoulli variable
x
t log ,
(161)
8
with the inverse:

4
1
1
1
x = 8et = 8 + 8t + 4t 2 + t 3 + t 4 + t 5 + t 6 + O t 7 .
(162)
3
3
15
90
Unlike the previous case (x = 0), the above series has an infinite radius of convergence:
r8 = .
(163)
65
Substituting the r.h.s. of the above equation in the series expansion for F1 (x) obtained previously
and, finally, expanding in powers of t, we have:

F1 (t) F1 x(t) =
n t n ,
(164)
n=0
where the first three coefficients n are:

1 = 8s0 ;
(165)
73
29 2
61
29
61
1
2 = a0 +
+
M0 2
M1 +
M2 2 log 2
log 2
54
5184
3456
576
864
48
29
1
+
(166)
log2 2 +
log 3;
576
64
343
203 2
929
203
929
3 =
a0

M0 2 +
M1
M2 2 log 2
486
46656
62208
5184
15552
7
7
203
+
(167)
log 2
log2 2
log 3.
192
5184
288
A computation of the convergence radius 8 of the series in t analogous to the one of the previous
section gives:
8 = log 2 0.693147.
(168)
7.2.4. The matching condition

a0 can be computed by imposing that the series in Eq. (101), which is completely determined,
and that one in Eq. (156) assume the same value in a given point in the intersection of the
respective domains of convergence. One has to take a point lying in the interval
1 < x < 1,
(169)
where the series of Eq. (101) converges, as well as in the interval

0 < x < 16,
(170)
where the series in Eq. (156) converges. One has therefore to choose a point in the interval
0 < x < 1,
(171)
such as for example x = 1/2. If we deal with infinite series, this procedure exactly determines
the coefficient a0 . As we have shown above, however, we can only determine an arbitrary, but
finite, number of coefficients of both series and the matching has to be made in an approximate
numerical way by using truncated series. The number of terms of the two series that must be
computed depend on the required precision on a0 . Our goal is to give F1 (x) with a relative precision of better then 1015 (double precision) using a relatively small number of terms in the
series (around 30). The problem is that any point in the interval (171) is close to the boundary
of the convergence domain for the series centered around x = 8, where convergence is slow, implying that many terms are needed for high accuracy. In other words, a direct matching between
the series in Eq. (101) and that one in Eq. (156) is not the good algorithm. As we have shown
before, a first improvement is obtained by re-writing the series expansions in terms of the relevant Bernoulli variables. But it is convenient to use additional series expansions centered around
auxiliary regular points in the range 0 < x < 8, such as for example x = 3. This topic will be
discussed in more detail in Section 7.5.
66
7.3. Expansion around x = large momentum expansion

In this section we consider the expansion of F1 (x) around x = , which generates a largemomentum expansion. Since the closest singularity to x = is at x = 8, we expect the expansion around infinity to be convergent outside the circle of radius 8, i.e. for
|x| > 8.
(172)
The expansion at infinity is studied systematically by changing variable to

1
x
and considering the limit
y
(173)
y 0.
(174)

Let us write the solution as usual as:
(0)
F1 (y) = S(y)
An y n .
(175)
n=0
The pre-factor is assumed to have power-like form,

S(y) = y ,
(176)
and to be a solution of the homogeneous equation in the limit (174):

3
4
S (y) S (y) + 2 S(y) = 0.
y
y
(177)
The indicial equation is ( 2)2 = 0, with a double zero in = 2 and therefore the solution is,
in the original x variable, of the form:
(0)
F1 (x) =

an
n=2
xn
+ log x

bn
n=2
xn
(178)
Because of the presence of the logarithmic term in front of the second series, the infinity is a
singular pointmore exactly, a branch point of infinite order. Furthermore, since positive powers of x do not appear, x = is a regular point for the series above. Let us write a bunch of
coefficients in terms of the lowest-order ones a2 and b2 :
a3 = 2a2 3b2 ;
(179)
b3 = 2b2 ;
33
a4 = 10a2 b2 ;
(180)
b4 = 10b2 ;
2
a5 = 56a2 100b2 ;
(181)
b5 = 56b2 ;
2561
a6 = 346a2
(182)
b6 = 346b2 ;
b2 ;
4
42631
b7 = 2252b2 .
b2 ;
a7 = 2252a2
(183)
10
Let us notice the power-like growth of the coefficients, which should scale asymptotically as
8n (b7 /b6 is already 6.5).
67

The non-trivial part of the expansion of the known term C(x) around x = is related to the
expansion of the GHPLs around this point, which is discussed in Appendix B. The expansion is
of the form:

kn
ln
mn
2
|x| > 8 ,
+
log
x
+
log
x
C(x) =
(184)
n
n
n
x
x
x
n=4
n=4
n=4
where the lowest-order coefficients read:

2
7
m4 = ;
;
l4 = 0;
(185)
48
16
1 13 2
7
43
;
l5 = ;
m5 = ;
k5 = +
(186)
2
48
4
16
1 125 2
131
331
;
l6 =
;
m6 =
;
k6 = +
(187)
8
48
8
16
40 357 2
3437
2569
;
l7 =
;
m7 =
;
k7 = +
(188)
9
16
24
16
18323 8827 2
56953
20301
+
;
l8 =
;
m8 =
.
k8 =
(189)
288
48
48
16
Since the differential equation involves a second derivative and the known term has doublelogarithmic terms, the solution must contain up to four powers of the logarithm:
k4 =
F1 (x) =

pn
n=2
xn
+ log x

qn
n=2
xn
+ log2 x

rn
sn
tn
3
4
+
log
x
+
log
x
.
xn
xn
xn
n=2
n=2
(190)
n=2
Substituting the above form for the solution into the equation, we obtain the following relations
among the coefficients:
r2 =
2
;
96
(191)
s2 = 0;
7
;
t2 =
192
(192)
(193)
5 2 27
(194)
+ ;
48
8
2 9
+ ;
q3 = 2q2 +
(195)
16 8
13 2
;
r3 =
(196)
16 48
7
s3 = ;
(197)
16
7
t3 = .
(198)
96
The values of p2 and q2 are left undetermined by the differential equation and a particular solution can be found by imposing for instance
p3 = 2p2 3q2
p2 = 0;
q2 = 0,
(199)
68
which we assume from now on. The general solution is given by:
F1 (x) = F1(0) (x) + F1 (x)
=

p n
n=2
xn
+ log x

qn
n=2
xn
+ log2 x

rn
sn
tn
3
4
+
log
x
+
log
x
,
xn
xn
xn
n=2
n=2
(200)
n=2
where we have defined:

p n pn + an ;
qn qn + bn .
(201)
Let us note that the coefficients of the double, triple and fourth logarithm are uniquely determined
because the associated homogeneous equation has no terms of this form: we therefore obtain
exact analytic expressions for these coefficients. On the contrary, the coefficients of the terms
with the single logarithm and of the terms without logarithm are determined up to a solution of
the homogeneous equation, and then they do depend on the two arbitrary constants
p 2
and q2 .
(202)
The latter can be found by matching the values of the series in Eq. (200) with that one in Eq. (156)
in a point in the range
8 < x < 16.
(203)
By following the matching procedure outlined in the previous section and described in detail in
Section 7.5, we obtain
p 2 = 1.04850063265766512303;
(204)
q2 = +1.50257112894949285675.
(205)
By decomposing the two-loop integral into different infrared and ultraviolet regions, a large
momentum expansion of F1 has been derived in Ref. [15], which reads in our normalization
containing an additional factor 1/16 :

1
2
5
31 4
7
1
4
2
log x
log x + (3) log x
+O 3 ,
F1 (x) = 2
(206)
96
4
2880
x 192
x

3
where (3)
n=1 1/n = 1.20206 . . . . By comparing with the first-order terms of our complete solution, we find an analytical agreement in the coefficient of the fourth, triple and double
logarithm.9 As far as the numerically determined coefficients are concerned, differences with the
analytical expressions are at most O(1019 ), well within the expected numerical uncertainty. We
can therefore use in place of our numerical estimates above the values:
31 4
5
(207)
q2 = + (3).
,
2880
4
Our final expression for the large-momentum expansion of the amplitude F1 (x) of the crossed
ladder diagram is that one in Eq. (200), with the first coefficients given by:
p 2 =
p 2 =
31 4
;
2880
(208)
9 We thank V. Smirnov for confirming a typo in the coefficient of the log2 x term in [15], which must be divided by
a factor 3.
69
5
q2 = + (3);
(209)
4
2
r2 = ;
(210)
96
s2 = 0;
(211)
7
t2 =
(212)
;
192
27
31 4 15
5
p 3 =
(213)
2
(3);
8
48
1440
4
9 2 5
+ (3);
q3 = +
(214)
8 16 2
13 2
;
r3 =
(215)
16 48
7
s3 = ;
(216)
16
7
t3 = .
(217)
96
Let us remark that Eqs. (213)(217) provide analytic expressions of the coefficients of the leading
power-suppressed terms in the large momentum expansion.
In order to accelerate the convergence of the series so far obtained, we move from the series
in x to the one in the Bernoulli variable
x8
,
t = log
(218)
x+1
with the inverse:

1 1 et
1
11 3
35 4
1733 5
7217 6
7 2
= t
=
t
t +
t +
t +
t + O t7 .
t
x 8+e
9
162
1458
52488
2361960
42515280
(219)
The convergence radius of the above series is equal to the one of the Bernoulli variable for the
expansion around x = 0:

r = 2 + 9 log2 2.
(220)
and finally expanding in powers of t , we have:
F1 (t) =

n=2
n t n + log x

n=2
n t n + log2 x
n t n + log3 x
n=2

n=2
n t n + log4 x
n t n ,
n=2
(221)
where the first three coefficients n , n , n , n , and n are:
1
a2 ;
81
1
2 = b2 ;
81
2 =
(222)
(223)
70
1
2;
7776
2 = 0;
7
2 =
;
15552
5
1
5
1
2
+
b2 +
a2 ;
3 =
34992
216 243
729
1
1
5
2
+
b2 ;
3 =
11664
648 729
5
13
2 +
;
3 =
69984
11664
7
;
3 =
11664
35
3 =
;
139968
95
2659
49
5
2
+
b2 +
a2 ;
4 =
1259712
839808 2187
26244
5
29
49
2
+
b2 ;
4 =
104976
46656 26244
49
125
2 +
;
4 =
2519424
139968
35
;
4 =
104976
343
,
4 =
5038848
where (see Eq. (199))
2 =
31 4
5
b2 = q2 = + z(3).
;
2880
4
The convergence radius of the expansion in t is:
a2 = p 2 =
= log
16
0.826679.
7
(224)
(225)
(226)
(227)
(228)
(229)
(230)
(231)
(232)
(233)
(234)
(235)
(236)
(237)
(238)
7.3.4. Analytic continuation to x =

The expansion of the amplitude F1 (x) for large time-like momenta, namely for (cf. Eq. (172))
< x < 8,
(239)
can be found from the asymptotic expansion in the space-like region (x > 8) simply by analytic
continuation. With the causal prescription
x |x| i,
(240)
we have in addition to the replacements (103):

log3 x log3 |x| 3 2 log |x| 3i log2 |x| + i 3 ;
log4 x log4 |x| 6 2 log2 |x| + 4 4i log3 |x| + 4i 3 log |x|.
(241)
71
F1 (x) then becomes complex for x < 0 with:

Re F1 (x) =

p n
n=2
xn
+ log |x|

qn
n=2
2
rn
2
+
log
|x|
xn
xn
n=2
3

sn
4
tn
2
2
2
4
+ log |x| 3 log |x|
+ log |x| 6 log |x| +
;
xn
xn
n=2
n=2
(242)

rn
sn
2
2
3
log
|x|
n
n
x
x
xn
n=2
n=2
n=2

tn
3
2
4 log |x| 4 log |x|
.
xn
Im F1 (x) =

qn
2 log |x|
(243)
n=2
7.4. Expansion around x = 1

In this section we consider the expansion around x = 1, the only singularity of the differential equation in the time-like region. Since the nearest singularity to x = 1 is the origin x = 0,
we expect the convergence radius of the expansion around this point to be
R1 = 1,
(244)
i.e. convergence in the circle

|x + 1| < 1.
(245)
On the real axis, that means:

2 < x < 0.
(246)

By solving the indicial equation, the solution turns out to be of the form:
(0)
F1 (x) =
an (x + 1)n + log(x + 1)
n=0
bn (x + 1)n .
(247)
n=0
The differential equation allows one to express all the coefficients of the expansion above in
terms of two of them, such as for example the lowest-order ones a0 and b0 :
4
4
a1 = a 0 + b 0 ;
3
9
41
62
a2 = a0 + b0 ;
27
81
400
2200
a3 =
a0 +
b0 ;
243
2187
4
b1 = b0 ;
3
41
b2 = b0 ;
27
400
b3 =
b0 .
243
(248)
(249)
(250)
We have a triangular structure of the coefficients, as in the case of the expansion around x = 0.
72

The expansion of the known term around x = 1 is of the form:
C(x) =
qn (x + 1)n + log(x + 1)
n=1
rn (x + 1)n .
(251)
n=0
The leading singularity for x 1 is therefore a single pole and the sub-leading singularity is a
logarithmic one related to the second series. The first few coefficients read:
5
5
2
+
K i
K ;
q1 =
(252)
24
24
12

19 2
101
1
101
7
+
5K + i
5K ;
q0 = +
(253)
4 216
1080
72 540
1
r0 = ;
(254)
4

2 131 2
1861
1627 1861
+
q1 = +
(255)
5K + i
5K ;
3 972
12150
6480 6075
47
r1 = ;
(256)
72

5177 1589 2
60178
129527 120356
q2 =
(257)
5K + i
5K ;
+
+
4320 8748
273375
291600 273375
749
r2 =
(258)
,
648
where K and K are two real transcendental constants defined as:
1
log y
2
1.97392;
dy =
5
y(y + 4)
0

1+ 5

K = log
0.481212.
2
(259)
(260)
Let us note that the coefficients above have an imaginary part related to the 2 massless particles
threshold in s = 0. Unlike previous cases, the expansion point lies indeed in the time-like region.
The particular solution of the inhomogeneous equation is of the same form as the homogeneous
one:
F1 (x) =
sn (x + 1)n + log(x + 1)
n=0
tn (x + 1)n .
(261)
n=0
For the coefficients, we find:

4
4
s1 = s0 + t0 + q1 ;
3
9
4
t1 = t0 ;
3

59 2
301
41
62
1
301
7
s2 = s0 + t0 +
+
5K + i
5 ;
27
81
8 864
4320
288 2160
(262)
(263)
(264)
73
1
41
(265)
+ t0 ;
16 27

400
2200
13
4381
313
169 2
4381
5K + i
5 ; (266)
s0 +
t0
+
+
s3 =
243
2187
54 1944
48600
6480 24300
1 400
t3 = +
(267)
t0 ;
8 243
etc., and we can choose s0 = t0 = 0. We assume therefore from now on the following expressions
of the coefficients:
t2 =
s0 = 0;
(268)
t0 = 0;
5
5
2
s1 =
+
K i
K ;
24
24
12
t1 = 0;

1
301
7
59 2
301
s2 = +
5K + i
5 ;
+
8 864
4320
288 2160
1
t2 = ;
16

13
4381
313
169 2
4381
5K + i
5 ;
+
s3 = +
54 1944
48600
6480 24300
1
t3 = .
8
Finally, the general solution is given by:
(0)
F1 (x) = F1 (x) + F1 (x)
sn (x + 1)n + log(x + 1)
n=0
(269)
(270)
(271)
(272)
(273)
(274)
(275)
(276)
tn (x + 1)n ,
(277)
n=0
where
and tn = bn + tn .
sn = an + sn
(278)
Eq. (277) depends on the two arbitrary constants,

a0
and b0 ,
(279)
that can be determined in a numerical way as shown above: we equate the series expansions
centered around x = 0 and x = 1, together with their derivatives, in one point where both
series converge, such as for example x = 1/2. That way we obtain:
a0 = 0.07572639563476980715 + i0.3122156449500221544;
|b0 | < 10
18
(280)
(281)
naturally implying
b0 = 0.
(282)
Let us observe that for x < 1 the logarithm above acquires an imaginary part related to the
3-particle threshold at s = m2 . Then, absorptive contributions related to both thresholds formally
74
come out as imaginary parts of the coefficients of the series and, via analytic continuation, of the
logarithmic pre-factor.
The vanishing of the coefficient b0 of the log(1 + x) term in the expansion of F1 (Eq. (282))
can be understood as follows. With a proper routing, F1 can be written as:

1
1
1
D 4 k1 D 4 k2
F1 =
2
2
2
2
2
(q + k1 ) + m (q + k2 ) + m (p1 + k1 ) (p2 + k2 )2
1
1
.
(283)
2
(p1 + k1 k2 ) (p2 k1 + k2 )2
By going to the threshold point q 2 = m2 and neglecting quadratic terms in the loop momenta
k12 , k22 , k1 k2 , one obtains in the soft limit:

1
1
1
1
1
1
. (284)
F1 D 4 k1 D 4 k2
q k1 q k2 p1 k1 p2 k2 p1 k1 p1 k2 p2 k1 + p2 k2
The integral above is convergent by power counting in the soft region. It is also easy to see that
there cannot be any collinear singularity. To have collinear singularities, one should have at least
one all massless 3-point vertex, i.e. a 3-point vertex connecting massless propagators and/or
external lines with light-cone momenta only, which is not the case. Since F1 has not any infrared
singularity in the point q 2 = m2 , it is finite for x 1 (F1 is ultraviolet finite, as shown
previously) and therefore cannot contain any log(1 + x) term in the expansion, implying b0 = 0.
In order to improve the convergence of the series so far obtained, we move from the series
in x to the one in the Bernoulli variable
t log(x),
(285)
with inverse:

1
1
1
1 5
1 6
t
t + O t7 .
x = et = 1 + t t 2 + t 3 t 4 +
2
6
24
120
720
The convergence radius of the series above is, as well known, infinite:
(286)
r1 = .
(287)
and finally expanding in powers of t , we have:

n t n + log(x + 1)
n t n ,
F1 (t) F1 x(t) =
n=0
(288)
n=0
where the first three coefficients n and n are:

0 = a 0 ;
5
5
4
1 = a0 +
+
K i
K ;
3
24
24
12

7
1
211
41 2
211
5K + i
5 ;
+
2 = +
8 864
4320
288 2160
The convergence radius of the expansion in t is given by:
2
1 = log 2 0.693147.
0 = 0;
(289)
1 = 0;
(290)
2 =
1
.
16
(291)
(292)
75
7.5. Series expansions around regular points and additional matching points
In the previous sections we illustrated the method of the solution of the differential equation
on the MI F1 (x), building series around the singular points only. That is certainly necessary but
not sufficient to determine F1 on the entire real axis. The expansions around all the singular
points do not allow indeed to compute the master integral in the range
8 < x < 2.
(293)
To cover this region, it is necessary to perform expansions of F1 (x) around auxiliary (regular)
points xi , expansions which are not multiplied by any singular function:
F1 (x) =
cn(i) (x xi )n .
(294)
n=0
It is clear that each new series expansion carries two unknown coefficients, which have to be
determined by matching with known series, as we have made for the singular points. To cover
the region (293), we have constructed two additional series expansions centered in
x = 8 and x = 3.
(295)
Even though the space-like region 0 < x < 8 is in principle covered by the expansions around
the singular points, we have found it convenient to consider also series expansions centered in
x=1
and x = 3.
(296)
The matching procedure is the following10 :

1. we match the series centered in x = 1 with the series centered in x = 0, which is completely
determined. That way we fix the two arbitrary coefficients entering the expansion in x = 1;
2. we then match the series in x = 3 with the one in x = 1;
3. we finally match the series in x = 8 with the series in x = 3.
With the above method, we have been able to cover the range 0 < x < 13 with a numerical
precision of about 21 digits. In the range 8 < x < , we need an additional series expansion in
x = 16, to be matched with the series centered in x = 8.
It is remarkable that the 9 expansion points above (4 singular + 5 regular ones) are sufficient
to build a routine evaluating numerically the function F1 (x) for every value of the variable x in
the real axis with a relative precision of about 1020 .
The internal consistency of the method can be checked as follows. The series expansion centered in x = 0 is completely determined by the initial conditions that we have been able to find.
By matching the free parameters of the series expansions centered in x = 1 and x = 1 with the
one centered in x = 0 we obtain two other series without free parameters. Now we move along
the real positive axis up to x = ;
x=0
x=1
x=3
x=8
x = .
(297)
10 The practical advantage of this method with respect to the general analytic continuation by (partially overlapping)
circles is related to the use of the differential equation to generate both series.
76
After that, we perform an analytic continuation in order to find the series expansion in x =
and, finally, we reach the series expansion in x = 1. What we find is a perfect agreement, within
the required precision, between the free parameters that we fixed matching the series expansion
in x = 1 with the one in x = 0 and those found performing the several matches along the real
axis, back to x = 1.11 Another (independent) check of our matching procedure is provided by
the values of the coefficients in Eqs. (204), (205), which are in agreement with the asymptotic
expansion of F1 (x) given in [15], as discussed before.
8. Expansions for the master integrals F2 and F3
The second MI F2 is directly determined from the first one by means of the (algebraic) equation (33) with i = 1, which gives
1 dF1
(x) + F1 (x),
F2 (x) = x
2 dx
(298)
in which F1 and F1 are assumed to be known. One just substitutes the power-series expansions
obtained so far for F1 (x) and obtains series representations for F2 (x) in the same convergence
domains.
The third MI satisfies a first-order equation of the form
1
dF3
(x) + F3 (x) + N (x) = 0,
dx
x
(299)
where the known term can be written as:

1
1
1
1
1
H (1, 0, 0; x) +
H (r, r, 0; x)
F1 (x)
1+
F2 (x) +
N(x) =
6x
3
x
96x 2
48x 2
+
2
H (1; x).
192x 2
(300)
It involves harmonic polylogarithms as well as F1 and F2 , which are assumed to be known.12

The associated homogeneous equation,
(0)
dF3
1 (0)
+ F3 (x) = 0,
dx
x
(301)
has the general solution

(0)
F3 (x) =
K
,
x
(302)
where K is an integration constant. A particular solution of the inhomogeneous first-order

equation is found in a straightforward manner with the general method of the variation of the
11 Since we deal in general with multivalued functions, one has to be careful to remain within the same sheet of the
Riemann surface by using consistently the causal +i prescription.
12 It is interesting to note that formally F and F appear as sub-topologies in the evaluation of F . That also implies
1
2
3
that the general method of series expansions allows to compute topologies with sub-topologies themselves given by series
expansions.
77
constants:
1
F3 (x) =
x
x
x N (x ) dx .
(303)
Since F1 and F2 are not given in closed form in terms of known transcendental functions, but
only in terms of (truncated) series expansions, one has also to expand all the GHPLs appearing
in N(x). As in the case of the first MI F1 , the general solution is:
K
(304)
+ F3 (x).
x
Analogously to the case of F1 , one can determine analytically all the coefficients of the expansion
of F3 around x = 0, by fixing
F3 (x) =
K =0
(305)
in order to avoid infrared power divergences. The matching procedure is similar to the one used
for F1 ; the main difference is that in this case there is only one undetermined coefficient for each
series because the differential equation is of first order. Details can be found in [22].
9. Relation to the equal-mass sunrise
There is a conspicuous relation between the homogeneous differential equation for the first
master integral F1 and the homogeneous equation for the equal mass sunrise in two dimensions,

1
S(z) =
(306)
D 2 k1 D 2 k2 ,
D1 D2 D3
where:
D1 = k12 + m2 ,
D2 = k22
(307)
+m ,
2
(308)
D3 = (p k1 k2 ) + m ,
2
(309)
and
s
,
(310)
m2
with s = p2 and p the external momentum. The homogeneous equation (see Eq. (4.3) of [13])
reads:

d2
1
1
1
d
1
1
1
S(z;
0)
+
+
+
S(z;
0)
+
S(z) = 0.
z z + 1 z + 9 dz
3z 4(z + 1) 12(z + 9)
dz2
(311)
By writing
z=
1
M(x 1),
x
whose inverse reads:
F1 (x) =
(312)
M(z) = (z + 1)F1 (z 1),
(313)
we find that M(z) satisfies the same homogeneous equation of S(z). The physical origin of
such relation is unclear to us.
78
10. Conclusions
We have computed the three master integrals for the crossed ladder diagram with two equalmass quanta exchanged by means of various power-series expansions centered around different
points. The two-equal-mass exchanged non-planar topology is by far the most complicated one
entering the two-loop form factor and a novel structure emerges, related to the occurrence of
elliptic integrals rather than (generalized) harmonic polylogarithms. That was not expected a priori, because the diagram (see Fig. 1) has only thresholds in s = 0 and in s = m2 , as well as a
pseudo-threshold in s = 4m2 ; we may only generically relate such structure to the non-planar
topology. We argue that the occurrence of elliptic integrals is probably a general feature of massive multi-loop diagrams beyond some level of complexity.
By studying the diagram close to the massless threshold s = 0 we have obtained a small
momentum expansion with coefficients analytically determined. By combining our largemomentum expansion with that obtained in Ref. [15], we have been able to give the leading power-suppressed corrections to the logarithmic contributions in a completely analytical
form.
We have found an a priori unexpected relation between the basic (first) master integral for
two-mass crossed ladder in four spacetime dimensions and the basic master integral for the
equal mass sunrise in two dimensions: the homogeneous differential equation is exactly the same,
while the inhomogeneous terms are unrelated (in the crossed ladder case, the latter ones being,
of course, much more complicated).
This work terminates the computation of the master integrals entering the two-loop electroweak form factor.
Acknowledgements
R.B. wishes to thank the Department of Physics of the University of Florence and INFN
Section of Florence for kind hospitality during a part of this work. The work of R.B. was partially
supported by Ministerio de Educacin y Ciencia (MEC) under grant FPA2004-00996, Generalitat
Valenciana under grant GV05-015, and MEC-INFN agreement.
Appendix A. Generalized harmonic polylogarithms
The general theory of the harmonic polylogarithms has already been discussed in [10,11,23],
to which we refer for details. Here we only give the specific information relevant to the computation of the crossed ladder. Let us first give the expressions of the harmonic polylogarithms
(HPLs) entering our computation in terms of the standard ones:
H (1; x) = log(1 + x);
(A.1)
1
log2 (x);
2
H (0, 1; x) = Li2 (x);
(A.3)
H (0, 1, 0; x) = 2 Li3 (x) log x Li2 (x).
(A.4)
H (0, 0; x) =
(A.2)
79
Let us now show that the Generalized Harmonic Polylogarithm (GHPL)13

x
log y
dy
y(4 y)
H (r, 0; x)
0
(A.7)
is real for 0 < x < 4, purely imaginary for x > 4, while it is complex for x < 0. In fact, for
0 < x < 4 the integrand is real in all the integrationdomain and therefore
also H is real. For
x > 4, since x x i with = +0 and, therefore, 4 x + i i x 4, one can split the

integral as:
4
H (r, 0; x) =
0
log y
dy i
y(4 y)
x
4
log y
dy.
y(y 4)
(A.8)
The first integral on the r.h.s. vanishes,

4
0
log y
dy = 0,
y(4 y)
(A.9)
implying that H is purely imaginary in this region. For x < 0 the integrand is complex and so is
the H . As a consequence, the term entering C(x), the inhomogeneous (and known) term in the
differential equation for F1 given in Eq. (54),
H (r, 0; x)
x(4 x)
(A.10)
is real for x > 0, as it should, while it is complex for x < 0.

Let us now show that H (r, r, 0; x) is real for x > 0 while it is complex for x < 0. Let us write:
x
H (r, r, 0; x)
0
1
H (r, 0; y) dy.
y(4 y)
(A.11)
For 0 < x < 4, the integrand is real in all the integration domain, while for x > 4 one can write:
4
H (r, r, 0; x) =
0
1
H (r, 0; y) dy i
y(4 y)
x
1
H (r, 0; y) dy.
y(y 4)
(A.12)
Both integrals on the r.h.s. are real.

13 The weight r labels the integration over the function
f (r; t) =
1
,
t (4 t)
(A.5)
such that, for instance:

x
H (r; x) =
0
dt
.
t (4 t)
(A.6)
80
H (r, 0; x) can be expressed in terms of ordinary harmonic polylogarithms of a non-linear

function of x, by changing the integration variable from y to

1
2
y
5
1
y
1 =y 2 2 3 +O 4 ,
t = 1+ y
(A.13)
2
4
y y
y
y
where the square root is the arithmetical one for y > 4, and on the last member we have made
the expansion for y 1. The inverse reads:
(t + 1)2
.
t
Note that y = 4 t = 1. We obtain:

x
x
1
H (r, 0; x) = 2iH 0, 1; 1 + x
2
4

2
x
x
1 +i .
+ iH 0, 0; 1 + x
2
4
6
y=
In a similar way, we obtain for the only GHPL of weight 3 the representation:

x
x
H (r, 0, 0; x) = H 0, 0, 0; 1 + x
1
2
4

x
x
1
+ 6 (3),
+ 4H 0, 1, 0; 1 + x
2
4
(A.14)
(A.15)
(A.16)
where H (0, 1, 0; 1) = 3/2 (3). In a completely analogous manner one can analyze H (r, r,
0; x), the GHPL entering the know term of the differential equation for F3 .
Appendix B. Expansion of generalized harmonic polylogarithms
The points x = 0, 4 and are singular points for H (r, 0; x). That is easily seen just by
looking at the integral definition. The Taylor expansion of H (r, 0; x) centered around a regular
point x0 = 0, 4, is easily obtained by remembering that
d
H (r, 0; x) = f (r, 0; x),
dx
(B.1)
log x
.
f (r, 0; x) =
x(4 x)
(B.2)
where
By performing the Taylor expansion of f (r, 0; x) and integrating on both sides, one obtains:
H (r, 0; x) = H (r, 0; x0 ) +

n=0
1
f (n) (r, 0; x0 )(x x0 )n+1 .
(n + 1)!
(B.3)
Let us note that one transcendental constant only is involved in this expansion (apart from trivial ones), namely H (r, 0; x0 ). Let us now consider the expansions around singular points. It is
clear that H (r, 0; x) does not possess an ordinary Taylor expansion around x = 0the origin is
81
a branch point of infinite orderwhere derivatives are not defined. This GHPL however has an
expansion around zero involving semi-integer powers of x multiplied with up to one power of
log x,
H (r, 0; x) = x 1/2
kn x n + x 1/2 log x
n=0
ln x n ,
(B.4)
n=0
where kn and ln are coefficients and the series on the r.h.s. have a convergence radius R = 4
i.e. up to the closest singularity. The representation above
can be obtained by expanding around
y = 0 the regular part of the integrand, i.e. the factor 1/ 4 y, so that

log y
1
3
5
1
log yy 1/2 +
log yy 3/2 +
log yy 5/2 + O y 7/2
= log yy 1/2 +
16
256
2048
y(4 y) 2
(B.5)
and using the result:

1
1
log yy dy =
(B.6)
y +1 .
log yy +1
+1
( + 1)2
The expansion around x = 4 is obtained in an analogous way, i.e. by expanding the regular part
of the integrand around x = 4, log y/ y, and integrating term by term. An expression of the
following form is obtained:
H (r, 0; x) = (x 4)1/2
gn (x 4)n ,
(B.7)
n=0
where
g0 = 2 log 2;
(B.8)
log 2
g1 =
(B.9)
;
12
1
3
log 2.
g2 =
(B.10)
80 320
x = 4 is therefore a branch point of second order. Let us note therefore that the expression (A.10)
has no singularities for x 4.
By using the representation obtained in the previous appendix, we can directly obtain the
expansion of the GHPL around infinity, which turns out to be of the form:
H (r, 0; x) = i

an
n=0
xn
i log x

bn
n=0
xn
1
i log2 x,
2
(B.11)
where the two series on the r.h.s. are convergent for |x| > 4 and the lowest-order coefficients
read:
2
;
6
a1 = 2;
3
a2 = ;
2
20
a3 = ;
9
a0 =
b0 = 0;
(B.12)
b1 = 2;
(B.13)
b2 = 3;
(B.14)
b3 =
20
.
3
(B.15)
82
References
[1] J.A. Aguilar-Saavedra, et al., ECFA/DESY LC Physics Working Group, hep-ph/0106315.
[2] J.A.M. Vermaseren, Symbolic Manipulation with FORM, Version 2, CAN, Amsterdam, 1991;
J.A.M. Vermaseren, New features of FORM, math-ph/0010025.
[3] Mathematica 5.2 (2005), a Mathematical Software System by Wolfram Research, Inc. (1987).
[4] G. Passarino, Nucl. Phys. B 619 (2001) 257, hep-ph/0108252;
G. Passarino, S. Uccirati, Nucl. Phys. B 629 (2002) 97, hep-ph/0112004;
G. Passarino, S. Uccirati, Nucl. Phys. B 747 (2006) 113, hep-ph/0603121;
A. Ferroglia, M. Passera, G. Passarino, S. Uccirati, Nucl. Phys. B 650 (2003) 162, hep-ph/0209219;
A. Ferroglia, M. Passera, G. Passarino, S. Uccirati, Nucl. Phys. B 680 (2004) 199, hep-ph/0311186;
S. Actis, A. Ferroglia, G. Passarino, M. Passera, S. Uccirati, Nucl. Phys. B 703 (2004) 3, hep-ph/0402132;
S. Actis, G. Passarino, S. Uccirati, Nucl. Phys. B (Proc. Suppl.) 160 (2006) 145, hep-ph/0608294;
T. Binoth, G. Heinrich, Nucl. Phys. B 585 (2000) 741, hep-ph/0004013;
Z. Nagy, D.E. Soper, Phys. Rev. D 74 (2006) 093006, hep-ph/0610028;
C. Anastasiou, A. Daleo, JHEP 0610 (2006) 031, hep-ph/0511176;
M. Czakon, Comput. Phys. Commun. 175 (2006) 559, hep-ph/0511200;
C. Anastasiou, S. Beerli, A. Daleo, hep-ph/0703282.
[5] M. Beccaria, G. Montagna, F. Piccinini, F.M. Renard, C. Verzegnassi, Phys. Rev. D 58 (1998) 093014, hep-ph/
9805250;
P. Ciafaloni, D. Comelli, Phys. Lett. B 446 (1999) 278, hep-ph/9809321.
[6] A. Denner, S. Pozzorini, Eur. Phys. J. C 18 (2001) 461, hep-ph/0010201;
A. Denner, S. Pozzorini, Eur. Phys. J. C 21 (2001) 63, hep-ph/0104127.
[7] B. Feucht, J.H. Kuhn, S. Moch, Phys. Lett. B 561 (2003) 111, hep-ph/0303016;
S. Pozzorini, Nucl. Phys. B 692 (2004) 135, hep-ph/0401087;
B. Feucht, J.H. Kuhn, A.A. Penin, V.A. Smirnov, Phys. Rev. Lett. 93 (2004) 101802, hep-ph/0404082;
A. Denner, S. Pozzorini, Nucl. Phys. B 717 (2005) 48, hep-ph/0408068;
B. Jantzen, J.H. Kuhn, A.A. Penin, V.A. Smirnov, Phys. Rev. D 72 (2005) 051301, hep-ph/0504111;
B. Jantzen, J.H. Kuhn, A.A. Penin, V.A. Smirnov, Phys. Rev. D 74 (2006) 019901, Erratum;
B. Jantzen, J.H. Kuhn, A.A. Penin, V.A. Smirnov, Nucl. Phys. B 731 (2005) 188, hep-ph/0509157;
B. Jantzen, J.H. Kuhn, A.A. Penin, V.A. Smirnov, Nucl. Phys. B 752 (2006) 327, Erratum;
B. Jantzen, V.A. Smirnov, Eur. Phys. J. C 47 (2006) 671, hep-ph/0603133;
A. Denner, B. Jantzen, S. Pozzorini, Nucl. Phys. B 761 (2007) 1, hep-ph/0608326.
[8] See for example: M. Ciafaloni, hep-ph/0612067.
[9] U. Aglietti, R. Bonciani, Nucl. Phys. B 668 (2003) 3, hep-ph/0304028.
[10] U. Aglietti, R. Bonciani, Nucl. Phys. B 698 (2004) 277, hep-ph/0401193.
[11] A.B. Goncharov, Math. Res. Lett. 5 (1998) 497516;
D.J. Broadhurst, Eur. Phys. J. C 8 (1999) 311, hep-th/9803091;
E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725, hep-ph/9905237;
T. Gehrmann, E. Remiddi, Nucl. Phys. B 601 (2001) 248, hep-ph/0008287;
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296, hep-ph/0107173;
T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 144 (2002) 200, hep-ph/0111255;
T.G. Birthwright, E.W.N. Glover, P. Marquard, JHEP 0409 (2004) 042, hep-ph/0407343;
D. Matre, Comput. Phys. Commun. 174 (2006) 222, hep-ph/0507152;
D. Matre, hep-ph/0703052;
J. Vollinga, S. Weinzierl, Comput. Phys. Commun. 167 (2005) 177, hep-ph/0410259;
S. Weinzierl, arXiv: 0705.0900 [hep-ph].
[12] For a general mathematical introduction, see for example: P. Dennery, A. Krzywicki, Mathematics for Physicists,
Dover, 1996;
or V. Smirnov, A Course of Higher Mathematics, vol. III, part 2, Pergamon Press, 1964.
[13] S. Laporta, E. Remiddi, Nucl. Phys. B 704 (2005) 349, hep-ph/0406160.
[14] S. Pozzorini, E. Remiddi, Comput. Phys. Commun. 175 (2006) 381, hep-ph/0505041.
[15] V.A. Smirnov, E.R. Rakhmetov, Theor. Math. Phys. 120 (1999) 870, hep-ph/9812529, Teor. Mat. Fiz. 120 (1999)
64.
83
[16] F.V. Tkachov, Phys. Lett. B 100 (1981) 65;

G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159;
T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485, hep-ph/9912329;
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 661 (2003) 289, hep-ph/0301170;
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 702 (2004) 359, Erratum.
[17] S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417;
S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033;
C. Anastasiou, A. Lazopoulos, JHEP 0407 (2004) 046, hep-ph/0404258.
[18] A.I. Davydychev, J.B. Tausk, Phys. Rev. D 53 (1996) 7381, hep-ph/9504431;
A.I. Davydychev, Phys. Rev. D 61 (2000) 087701, hep-ph/9910224;
R. Bonciani, P. Mastrolia, E. Remiddi, Nucl. Phys. B 690 (2004) 138, hep-ph/0311145;
R. Scharf, J.B. Tausk, Nucl. Phys. B 412 (1994) 523;
F.A. Berends, A.I. Davydychev, V.A. Smirnov, J.B. Tausk, Nucl. Phys. B 439 (1995) 536, hep-ph/9410232;
J. Fleischer, M.Y. Kalmykov, A.V. Kotikov, Phys. Lett. B 462 (1999) 169, hep-ph/9905249;
J. Fleischer, A.V. Kotikov, O.L. Veretin, Nucl. Phys. B 547 (1999) 343, hep-ph/9808242;
J. Fleischer, V.A. Smirnov, O.V. Tarasov, Z. Phys. C 74 (1997) 379, hep-ph/9605392.
[19] A.V. Kotikov, Phys. Lett. B 254 (1991) 158;
A.V. Kotikov, Phys. Lett. B 259 (1991) 314;
A.V. Kotikov, Phys. Lett. B 267 (1991) 123;
E. Remiddi, Nuovo Cimento A 110 (1997) 1435, hep-th/9711188;
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Acta Phys. Pol. B 29 (1998) 2627, hep-th/9807119;
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365, hep-th/9805118.
[20] G. t Hooft, M.J.G. Veltman, Nucl. Phys. B 153 (1979) 365;
L.C. Maximon, Proc. R. Soc. London, Sect. A 459 (2003) 2807.
[21] See for example: A. Markusevic, The Theory of Analytic Functions: A Brief Course, Mir Publishers, Moscow,
1983.
[22] L. Grassi, Un nuovo metodo per il calcolo numerico preciso di diagrammi di Feynman, Tesi di Laurea in Fisica
Teorica, Universita di Bologna, A.A. 20042005, Master Thesis (in Italian).
[23] U. Aglietti, R. Bonciani, G. Degrassi, A. Vicini, Phys. Lett. B 595 (2004) 432, hep-ph/0404071;
U. Aglietti, R. Bonciani, G. Degrassi, A. Vicini, Phys. Lett. B 600 (2004) 57, hep-ph/0407162.
Non-Abelian vortices and monopoles in SO(N) theories

L. Ferretti a,b , S.B. Gudnason c,d , K. Konishi c,d,
a SISSA, via Beirut 2-4, I-34100 Trieste, Italy
b INFN, Sezione di Trieste, I-34012 Trieste (Padriciano), Italy
c Department of Physics, E. Fermi, University of Pisa, Largo Pontecorvo 3, Ed. C, 56127 Pisa, Italy
d INFN, Sezione di Pisa, Largo Pontecorvo 3, Ed. C, 56127 Pisa, Italy
Received 9 July 2007; accepted 20 July 2007

Abstract
Non-Abelian BPS vortex solutions are constructed in N = 2 theories with gauge groups SO(N ) U (1).
The model has Nf flavors of chiral multiplets in the vector representation of SO(N ), and we consider a color-flavor locked vacuum in which the gauge symmetry is completely broken, leaving a global
SO(N )C+F diagonal symmetry unbroken. Individual vortices break this symmetry, acquiring continuous
non-Abelian orientational moduli. By embedding this model in high-energy theories with a hierarchical
symmetry breaking pattern such as SO(N + 2) SO(N ) U (1) 1, the correspondence between nonAbelian monopoles and vortices can be established through homotopy maps and flux matching, generalizing
the known results in SU(N ) theories. We find some interesting hints about the dual (non-Abelian) transformation properties among the monopoles.
1. Introduction
Recently some significant steps have been made in understanding the non-Abelian monopoles
[18], occurring in spontaneously broken gauge field theories [9,10]. The basic observation is
that the regular t HooftPolyakov-like magnetic monopoles occurring in a system
v1
GH,
(1.1)
* Corresponding author at: Department of Physics, E. Fermi, University of Pisa, Largo Pontecorvo 3, Ed. C, 56127
Pisa, Italy.
E-mail addresses: ferretti@sissa.it (L. Ferretti), gudnason@df.unipi.it (S.B. Gudnason), konishi@df.unipi.it
(K. Konishi).
doi:10.1016/j.nuclphysb.2007.07.021
L. Ferretti et al. / Nuclear Physics B 789 (2008) 84110
85
where H is a non-Abelian unbroken gauge group, are not objects which transform among
themselves under the unbroken group H , but which transform, if any, under the magnetic dual
of H , namely H . As field transformation groups, H and H are relatively non-local, thus a local
transformation in the magnetic group H would look like a non-local transformation in the electric
theory. Although this was implicit in the work by GoddardNuytsOlive [2] and others [3,4], the
lack of the concrete knowledge on how H acts on semiclassical monopoles has led to longstanding puzzles and apparent difficulties [6,7].
Detailed study of gauge theories with N = 1 or N = 2 supersymmetry and quark multiplets,
on the other hand, shows that light monopoles transforming as multiplets of non-Abelian magnetic gauge group H do occur quite regularly in full quantum systems [1114]. They occur under
certain conditions, e.g., that there is a sufficiently large exact flavor symmetry group in the underlying theory, which dresses the monopoles with flavor quantum numbers, preventing them from
interacting too strongly. Also, the symmetry requirement (i.e., the symmetry of the low-energy
effective theory describing the light monopoles be the correct symmetry of the underlying theory) seems to play an important role in determining the low-energy degrees of freedom in each
system [15]. There are subtle, but perfectly clear, logical reasons behind these quantum mechanical realizations of dual gauge symmetries in supersymmetric models. Since there are free
parameters in these supersymmetric theories which allow us to move from the fully dynamical
regime to semiclassical regions, without qualitatively changing any physics, it must be possible
to understand these light degrees of freedom in terms of more familiar soliton-like objects, e.g.,
semiclassical monopoles.
This line of thought has led us to study the system (1.1), in a regime of hierarchically broken
gauge symmetries
v1
v2
GH 1,
v1 v 2 ,
(1.2)
namely, in a phase in which the unbroken H gauge system is completely broken at much
lower energies (Higgs phase), so that one expectsbased on the standard electromagnetic duality
argumentthe H system to be in confinement phase. The elementary monopoles confined by
the confining strings in the H theory should look like t HooftPolyakov monopoles embedded
in a larger picture where their magnetic fluxes are frisked away by a magnetic vortex of the H
theory in Higgs phase.
Indeed, in the context of softly broken N = 2 models, this kind of systems can be realized
concretely, by tuning certain free parameters in the models, typically, by taking the bare quark
masses m (which fix the adjoint scalar VEVs, = v1 m) much larger than the bare adjoint
scalar mass (which sets the scale for the squark VEVs, q = v2 m ). In a high-energy
approximation, where v2 is negligible, one has a system, (1.1), with a set of t HooftPolyakov
monopoles. In the class of supersymmetric models considered, these monopoles are BPS, and
their (semiclassical) properties are well understood. In the low-energy approximation (where the
massive monopoles are integrated out and v1 is regarded as infinitely large) one has the H theory
in Higgs phase, with BPS vortices whose properties can also be studied in great detail.
When the full theory is considered, with small corrections which involves factors of vv21 ,
there is an important qualitative change to be taken into account at the two sides of the mass
scales (high-energy and low-energy). Neither monopoles of the high-energy approximation nor
the vortices of the low-energy theory, are BPS saturated any longer. They are no longer topologically stable. This indeed follows from the fact that 2 (G) is trivial for any Lie group (no
regular monopoles if H is completely broken) or if 1 (G) = 1 (there cannot be vortices). If
86
1 (G) = 1 there may be some stable vortices left, but still there will be much fewer stable vortices as compared to what is expected in the low-energy theory (which sees only 1 (H )). As
the two effective theories must be, in some sense, good approximations as long as vv21
1, one
faces an apparent paradox.
The resolution of this paradox is both natural and useful. The regular monopoles are actually
sources (or sinks) of the vortices seen as stable solitons in the low-energy theory; vice versa, the
vortices which should not be there in the full theory, simply end at a regular monopole. They
both disappear from the spectrum of the respective effective theories. This connection, however,
establishes one-to-one correspondence between a regular monopole solution of the high-energy
theory and the appropriate vortex of the low-energy theory. As the vortex moduli and non-Abelian
transformation properties among the vortices, really depend on the exact global symmetry of the
full theory (and its breaking by the solitons), such a correspondence provides us with a precious
hint about the nature of the non-Abelian monopoles. In other words, the idea is to make use of
the better understood non-Abelian vortices to infer precise conclusions about the non-Abelian
monopoles, by-passing the difficulties associated with the latter as mentioned earlier.
A quantitative formulation of these ideas requires a concrete knowledge of the vortex moduli space and the transformation properties among the vortices [1618]. This problem has
been largely clarified, thanks to our generally improved understanding of non-Abelian vortices
[1923], and in particular to the technique of the moduli matrix [24], especially in the context
of SU(N ) gauge theories. Also, some puzzles related to the systems with symmetry breaking
SO(2N) U (N ), or SO(2N ) U (r) U (1)Nr , have found natural solutions [9].
In this article, we wish to extend these analyses to the cases involving vortices of SO(N )
theories. In [25] the first attempts have been made in this direction, where softly broken N = 2
models with SO(N ) gauge groups and with a set of quark matter in the vector representation, have
been analyzed. In the case of SO(2N + 3) theory broken to SO(2N + 1) U (1) (with the latter
completely broken at lower energies) one observes some hints how the dual, USp(2N ) group,
might emerge. In the model considered in [25], however, the construction of the system in which
the gauge symmetry is completely broken, leaving a maximum exact color-flavor symmetry (the
color-flavor locking), required an ad hoc addition of an N = 1 superpotential, in contrast to
SU(N ) theories where, due to the vacuum alignment with bare quark masses familiar from N = 1
SQCD, the color-flavor locked vacuum appears quite automatically.
In this article we therefore turn to a slightly different class of SO(N ) models. The underlying
theory is an SO(N + 2) gauge theory with matter hypermultiplets in the adjoint representation,
with the gauge group broken partially at a mass scale v1 . The analysis is slightly more complicated than the models considered in [25], but in the present model the color-flavor locked vacua
occur naturally. Also, these models have a richer spectrum of vortices and monopoles than in the
case of [25], providing us with a finer testing ground for duality and confinement.
At scales much lower than v1 , the model reduces to an SO(N ) U (1) theory with quarks in
the vector representation. Non-Abelian vortices arising in the color-flavor locked vacuum of this
theory transform non-trivially under the SO(N )C+F symmetry. We are interested in their role in
the dynamics of gauge theories, but these solitons also play a role in cosmology and condensed
matter physics, so the results of Sections 3 and 4 of this paper could be of more general interest
(for example they can be useful for cosmic strings, see [29]).
In Section 2 of this article, we present the high-energy model with gauge group SO(2N + 2).
In Section 3 we study its low-energy effective theory and present the vortex solutions. In Section 4 we study the model with gauge group SO(2N + 3). Finally, in Section 5 we discuss the
correspondence between monopoles and vortices.
87
2. The model
We shall first discuss the SO(2N + 2) theory; the case of SO(2N + 3) group will be considered
separately later. We wish to study the properties of monopoles and vortices occurring in the
system
v1
v2
SO(2N + 2) SO(2N ) U (1) 1.
(2.1)
To study the consequences of such a breaking, we take a concrete example of an N = 2 supersymmetric theory with gauge group SO(2N + 2) and Nf matter hypermultiplets in the adjoint
representation. All the matter fields have a common mass m, so the theory has a global U (Nf )
flavor symmetry. We also add a small superpotential term 2 in the Lagrangian, which breaks
softly N = 2 to N = 1. For the purpose of considering hierarchical symmetry breaking (2.1), we
take
m .
(2.2)
The theory is infrared-free for Nf > 1, but one may consider it as an effective low-energy theory
of some underlying theory, valid at mass scales below a given ultraviolet cutoff. In any case, our
analysis will focus on the questions how the properties of the semiclassical monopoles arising
from the intermediate-scale can be understood through the moduli of the non-Abelian vortices
arising when the low-energy, SO(2N ) theory is put in the Higgs phase.
The superpotential of the theory has the form,

Tr qA [, qA ] + m
Tr qA qA + Tr 2 .
W= 2
(2.3)
2
A
The vacuum equations for this theory therefore take the form

, = 0,

qA , qA =
qA , qA ,
A
(2.4)
(2.5)

2[qA , qA ] + = 0,
(2.6)
2[, qA ] + mqA = 0,
2[, qA ] + mqA = 0.
We shall choose a vacuum in which takes the vacuum expectation value (VEV)
0 iv 0 0
iv
0
0 0
0
0
0 0,
=
.
..
.. . .
..
. 0
.
.
0
0
0 0 0
(2.7)
(2.8)
(2.9)
which breaks SO(2N + 2) to SO(2N ) U (1) and is consistent with Eq. (2.4).
We are interested in the Higgs phase of the theory. In order for the SO(2N ) U (1) symmetry
to be broken at energies much lower than v1 v, we have to find non-vanishing VEVs of the
squarks which satisfy Eqs. (2.7), (2.8). This means that v O(m). The magnitude of squark
VEVs is then fixed by Eq. (2.6) to be of the order of (m)1/2

m and defining v2 m
88
we obtain the hierarchical breaking of the gauge group (2.1). The D-term condition (2.6) can be
satisfied by the ansatz
q = q .
(2.10)
One must also determine the components of the fields q, q which do not get a mass of the order of O(v) O(m). We see from Eq. (2.3) that the light squarks are precisely those for which
Eqs. (2.7), (2.8) are satisfied non-trivially, i.e., by non-vanishing eigenvectors q, q.
The conditions (2.7), (2.8) require that the light components correspond to the generators of SO(2N ) which
are lowering and raising operators for . This condition implies also
m
v= .
2
(2.11)
To find the light components of q, q,

we note that for a single flavor, Eqs. (2.6)(2.8) together
have the form of an su(2) or su(3) algebra, T1 , T2 , T3 ,
T3 ,
qA T = T1 iT2 ,
qA T+ = T1 + iT2 ,
(2.12)
with appropriate constants.

The simplest way to proceed is to consider the various SO(3) subgroups, SO(3)12j , lying in
the (12j ) three-dimensional subspaces (j = 3, 4, 5, . . .), with
0 i 0
T3 = H (0) = i12 = i 0 0 ,
(2.13)
0 0 0 12j
0 0 1
T+ = T1 + iT2 = Lj, .
T = T1 iT2 = Lj, 0 0 i ,
(2.14)
1 i 0 12j
The light fields which remain massless can then be expanded as
qA (x) =

j =3,4,5,...
1
qj A (x)Lj, ,
2
qA (x) =

j =3,4,5,...
1
qAj (x)Lj,+
2
for each flavor A = 1, 2, . . . , Nf . Written as a full SO(2N ) matrix, Lj, looks like
0
0 ... 1 ...
..
0
0
i
.
..
.
..
,
.
Lj, =
Lj,+ = Lj, .
..
1 i
.
..
.
...
... 0
(2.15)
(2.16)
In Lj, the only non-zero elements (1 and i) in the first two rows appear in the (2 + j )th
column; the only two non-zero elements in the first two columns (1 and i) appear in the (2 +
j )th row.
An alternative way to find the combinations which do not get mass from is to use the
independent SU(2) subgroups contained in various SO(4) subgroups living in the subspaces
(1, 2, j, j + 1), j = 3, 5, . . . , 2N 1. As is well known, the so(4) algebra factorizes into two
89
commuting su(2) algebras,

so(4) su(2) su(2),
(2.17)
where for instance for SO(4)1234 one has

i
S1 = (23 + 41 ),
2
i
S1 = (23 41 ),
2
where

0 1
,
23 =
1 0 23
i
S2 = (31 + 42 ),
2
i
S2 = (31 42 ),
2
is (up to a phase) the rotation generator in the 23 plane, etc.

Since
2
= H (0) = i12 = S3 + S3 ,
m
i
S3 = (12 + 43 ),
2
i
S3 = (12 43 ),
2
(2.18)
(2.19)
(2.20)
it follows from the standard su(2) algebra that both S = S1 iS2 and S = S1 i S2 satisfy the
relation,

2
2
(2.21)
, S = S ,
, S = S .
m
m
One can choose the two combinations
L = S + S ,
L = S S ,
which satisfy the required relation,

2
2

, L = L ,
, L = L .
m
m
(2.22)
(2.23)
These constructions can be done in all su(2) subalgebras living in SO(4)1,2,j,j +1 , j = 3, 5,

. . . , 2N 1.
Explicitly, Sj , Sj , and Lj, , Lj, have the form (j = 3, 5, . . .)
0
0
1 i
1 0
0 i 1
Sj =
,
0 0
2 1 i
i 1 0 0 (1,2,j,j +1)
0 0 1 i
1 0 0 i 1
,
Sj =
(2.24)
0
2 1 i 0
i
1 0
0 (1,2,j,j +1)
0 0 1 0
0 0 i 0
Lj, =
,
1 i 0 0
0 0 0 0 (1,2,j,j +1)
90
0
Lj, =
0
i
0 0 i
0 0 1
.
0 0 0
1 0 0 (1,2,j,j +1)
(2.25)
Clearly, one can write

Lj, = iLj +1, ;
(2.26)
and use the first of Eq. (2.25) to define Lj, for all j = 3, 4, 5, . . . , j even or odd. With this
definition, Lj, coincide with those introduced in Eq. (2.14) by using various SO(3) subgroups.
Eqs. (2.3), (2.21), (2.23) show that the light fields (those which do not get mass of order m)
are the ones appearing in the expansion (2.15). Alternatively, the basis of light fields can be taken
as

1
iA (x)Si, ,
QiA (x)Si, + Q
qA (x) =
2 i=3,5,...
1
(x)S .
Q A =
(2.27)
QAi (x)Si,+ + Q
Ai
i,+
2 i=3,5,...
The relation between the qiA (x) and QiA (x) fields is (i = 3, 5, . . .):
qiA (x) + iqi+1,A (x)
;
2
qA,i (x) iqA,i+1 (x)
Q iA (x) =
(2.28)
= Qi+1,A (x).
2
All other components get a mass of order m. There are thus precisely 2N light quark fields
(color components) qiA (i = 1, 2, . . . , 2N ) for each flavor. These are the light hypermultiplets of
the theory.
Each of the two bases {qiA } or {QiA } has some advantages. Clearly the basis qiA (i =
1, 2, . . . , 2N ) corresponds to the usual basis of the fundamental (vector) representation of
the SO(M) group (M = 2N ), appearing in the decomposition of an adjoint representation of
SO(M + 2) into the irreps of SO(M):
QiA (x) =
(M + 2)(M + 1) M(M 1)
=
M M 1.
(2.29)
2
2
The low-energy effective Lagrangian can be most easily written down in terms of these fields,
and the symmetry property of the vacuum is manifest here.
On the other hand, the basis (Qj A , Q j A ), j = 3, 5, 7, . . . , is made of pairs of eigenstates of
the (a (j 1)/2)th Cartan subalgebra generator,
j 1
(2.30)
= 1, 2, . . . , N
2
(see Eqs. (2.18)(2.20)), with eigenvalues 1, so that the vortex equations can be better formulated, and the symmetry maintained by individual vortex solutions can be seen explicitly in
In
iA (i = 3, 5, . . .), form an N.
this basis. QiA (i = 3, 5, . . .), form an N of SU(N ) SO(2N ); Q
other words, it represents the decomposition of a 2N of SO(2N ) into N+ N of SU(N ) SO(2N ).

is discussed more extenThe change of basis from the vector basis (q) and U (N ) basis (Q, Q)
sively in Appendix A.
H (a) = ij,j +1 = Sj,3 Sj,3 ,
a=
91
3. Vortices in the SO(2N) U (1) theory

3.1. The vacuum and BPS vortices
The low-energy Lagrangian for the theory with gauge group SO(2N ) U (1) and squarks
qA qA in the fundamental representation of SO(2N ) is
L=

2
1
1 0 0
b
F F 2 F b F
+ |D qA |2 + D qA
2
4g1
4g2N
2

g2N
q t b qA qA t b q 2 2g 2 qA t b qA 2
2N
A
A
2

g12 0
m 2
0 2
2
q t qA qA t qA 2g1 qA qA + + ,
2 A
2
(3.1)
where the dots denote higher orders in /m and terms involving = . Note that to this
order the only modification is a FayetIliopoulos term which does not break N = 2 SUSY. The
covariant derivative acts as
D qA = qA iA0 qA iAb t b qA ,
(3.2)
where t a are normalized as

2
Tr t a = 1,
(3.3)
and
1
1
t = H (a) =
2
2
a
0
i
i
0
(3.4)
2a+1,2a+2
where H (a) is the ath Cartan generator of SO(2N ), a = 1, 2, . . . , N , which we take simply as

0 i
(a)
H =
(3.5)
.
i 0 2a+1,2a+2
As we have seen already, each light field carries unit charge with respect to H (0) ; the pair
A,j ), j = 3, 5, 7, . . . , furthermore carries the charge 1 with respect to H (a) (a =
(QA,j , Q
(j 1)/2) and zero charge with respect to other Cartan generators.
Let us define = m/2 which is the only relevant dimensional parameter in the Lagrangian.
We set Nf = 2N , which is enough for our purposes.1 By writing qiA , qAi as color-flavor mixed
matrices q, q,
the vacuum equations are now cast into the form

Tr qq = Tr q q ,
T

T
qq qq = q q q q ,
(3.6)
(3.7)
1 Higher N are interesting because of semilocal vortex configurations arising in this theories. These solutions will be
f
discussed elsewhere.
92
Tr(q q)
+ = 0,
2
2
b
Tr t q q = 0.
(3.8)
(3.9)
The vacuum we choose to study is characterized by the color-flavor locked phase

v2 =
qA,j = qA,j = A,j v2 ,
,
2N
or
1 0 0 0
0 1 0 0

q = q = v2 1 = v2
,
0 0 ... 0
0 0
(3.10)
(3.11)
which clearly satisfies all the equations above. The gauge (O) and flavor (U ) transformations act
on them as
q OqU T ,
q U qO
T,
O SO(2N ) U (1),
U U (2N ),
(3.12)
the gauge group is completely broken, while a global SO(2N )C+F U (1)C+F group (U = O)
is left unbroken.
When looking for vortex solutions, one suppresses time and z dependence of the fields and
retains only the component Fxy of the field strength. The vortex tension can be cast in the Bogomolnyi form:

2

1
1 0

2
T = d 2 x
Fijb g2N ij qA t b qA +
Fij g1 ij qA qA
2g2N
2g1

2

1
1
2
+ Di qA iij Dj qA + Di qA iij Dj qA
2
2

2

g2N
q t b qA qA t b q 2 + g 2 q qA qA q 2 ij F 0 .
+
(3.13)
1 A
ij
A
A
2 A
The terms with the square brackets in the last line of Eq. (3.13) automatically vanish with the
ansatz [20]
qiA = qiA
,
(3.14)
thus we shall use this ansatz for the vortex configurations. The resulting BPS equations are

1 0
Fij + g1 ij qA qA = 0,
2g1
1
F b + g2N ij qA t b qA = 0,
2g2N ij
Di qA + iij Dj qA = 0, = 1,
where we have used the ansatz (3.14). The tension for a BPS solution is

T = d 2 x ij Fij0 .
(3.15)
(3.16)
(3.17)
(3.18)
93
To obtain a solution of these equations, we need an ansatz for the squark fields. It is convenient
to perform a U (2N)F transformation (2.28), where the vacuum takes the block-diagonal form
1 1 0 0

i i 0 0

1
0 0 1 1 .
Q = Q =
(3.19)

2N
2 0 0 i i
..
..
..
.. . .
.
.
.
.
.
In this basis, the ansatz is:
rj
1
;
t0 ;
2
r
2
in+ +
1
1 (r)
ein1 1 (r)
e
iein+1 + (r) iein1 (r)
1
1
1
0
0
Q(r, ) =
2
0
0
..
..
.
.
Ai = ha (r)t a ij
(3.20)
0
0
+
ein2 2+ (r)
+
iein2 2+ (r)
..
.
0
0
ein2 2 (r)
iein2 2 (r)
..
.
..
.
(3.21)
where t a s are the generators of the Cartan subalgebra of SO(2N ). The conditions for the fields
at r are fixed by the requirement of finite energy configurations:

,
a () =
(3.22)
2N
1
1
(0)
n(a) ,
n(0) h0 (),
n(a) ha (),
n
(3.23)
a =n
2
2
where n(0) and n(a) are the winding numbers with respect to the U (1) and to the ath Cartan
U (1) SO(2N) defined in Eq. (3.5).
Clearly
(0)
N 0 n+
a + na = 2n ,
(3.24)
is independent of a. The regularity of the fields requires that the QA s come back to their original
value after a 2 rotation, and this yields the quantization condition,
n
a Z,
a,
(3.25)
implying that the U (1) winding numbers n(0) and n(a) are quantized in half integer units, consistently with considerations based on the fundamental groups (see Appendix B and below).
We need only the information contained in Eqs. (3.20), (3.23) to evaluate the tension for a
BPS solution:

d rA0 (r) = 2 2 h0 () = 2 N0 = 2 |N0 |.

T = 2 lim
(3.26)
r
The last equality comes from the requirement for the tension to be positive, so = sign(N0 ).
Note that the tension depends only on |N0 |, which is twice the U (1) winding.
94
From the BPS equations we obtain the differential equations for the profile functions h0 ,
ha , a :

+ 2 2
dh0
+ ,
= 2 2g12 r
a
a
dr
a
(3.27)
2 2
dha
2
r a+ a ,
= 2 2g2N
dr

da
h0 ha a
= na
.
dr
r
2
(3.28)
(3.29)
N0
and fa = ha +
In order to cast them in a simple form, we define f0 = h0
2
n+
n
a
a
2
+ 2 2
df0
+ ,
= 2 2g12 r
a
a
dr
a
and obtain
(3.30)
2 2
dfa
2
r a+ a ,
= 2 2g2N
dr

da
f0 fa a
=
.
dr
r
2
(3.31)
(3.32)
The boundary conditions at r are

a () =
,
2N
f0 () = fa () = 0,
(3.33)
There are also regularity conditions at r = 0 for the gauge fields h0 (0) = ha (0) = 0 which are
N0
f0 (0) = ,
2
fa (0) =
n+
a na
.
(3.34)
Solving Eq. (3.32) for small r with the conditions (3.34), we obtain a r na . To avoid a
singular behavior for these profile functions we need

sign n
a = .
(3.35)
This condition is consistent with = sign(N0 ). With this condition there are no singularities
at r = 0 and Eqs. (3.30), (3.31), (3.32) can be solved numerically with boundary conditions
(3.33), (3.34).
The profile functions for the simplest vortex N0 = 1, n+

1 = 1, n1 = 0 in the SO(2) U (1) the+
ory are shown in Figs. 1, 2. The profile functions (f0 , fa , a , a ) for the minimal vortex N0 = 1,
2
2
n+
i = 1, ni = 0 in the SO(2N ) U (1) theory can be obtained by rescaling g2N g
2N /N and
then taking all a equal to the profile functions shown above rescaled by a factor 1/ N . Sim
ilarly, solutions corresponding to the exchange (n+
a , na ) = (1, 0) (0, 1) can be obtainedby
+
exchanging fa fa and a a . The typical length scale of the profile functions is 1/ ,

which is the only dimensional parameter in the Bogomolnyi equations.
95
Fig. 1. Numerically integrated

minimum vortex solution with N0 = 1, where we have taken the couplings to be
2 = 1. ( 2f ).
4g12 = 4g2N
i
i
Fig. 2. Numerically integrated minimum vortex solution with N0 = 1, where we have taken the couplings to be 4g12 = 1
2 = 2 for the left panel and 4g 2 = 2 and 4g 2 = 1 for the right panel. ( 2f ).
and 4g2N
i
i
1
2N
3.2. Vortex moduli space

To study the space of solutions of the BPS equations we have obtained above, it is convenient
to rewrite the ansatz (3.21) for the squark fields in the original basis:
0
0
M1 (r, )
0
0
M2 (r, )
q(r, ) =
0
0
M3 (r, ) ,
..
..
..
..
.
.
.
.
in+ +

+
in
a
a
1
e
a (r) + e
a (r)
i(eina a+ (r) eina a (r))
. (3.36)
Ma (r, ) =
+
eina a+ (r) + eina a (r)

2 i(eina a+ (r) eina a (r))
In this basis the action of the SO(2N )C+F transformations on squark fields is simply q =
) is a solution to the BPS equations, O q(r,
)O T is
OqO T . The first observation is that if q(r,
also a solution. Note also that these solutions are physically distinct because they are related by
a global symmetry. In this way, from a single solution of the form (3.21), we can obtain a whole
96
continuous SO(2N ) orbit of solutions. Any given vortex solution is a point in the moduli space
and SO(2N)C+F acts as an isometry on this space.
From Eqs. (3.23) and (3.35), we see that regular solutions are described by a set of 2N + 1
integers N0 n
a which satisfy the following conditions:
n+
a + na = N0 , a,
+

sign na = sign n
a = sign(N0 ),
(3.37)
a,
(3.38)
where N0 Z is related to the winding around the U (1) and is the only parameter of the solution
which enters the tension T = 2 |N0 |.
Let us study the solutions with the minimum tension. Minimal vortices have N0 = 1 and
T = 2 . Note that solutions with N0 < 0 can be obtained by taking the complex conjugate of
solutions with N0 > 0, so from now on we will consider only solutions with positive N0 . These
vortices can be divided into two groups, the first has 2N1 representative (basis) vortices which
are

+

n
n1
1 0
0 1
1

n+
n2 1 0 0 1
2

..
.. =
... ... , 1 0 , . . . ,
N0 = 1,
(3.39)
.
.
. .
+
1 0 .. ..
n
N1 nN1
1 0
1 0
n+
n
N
N
which all have an even number of n
i s equal to 1; and
vortices, characterized by the integers

+

n
n1
1 0
1
1

..
n+
n
1
0
2
2
.
.
.
..
..
.. .. , 1
N0 = 1,
=
.
.

+
1 0 0
n
n
N1
N1
0 1
1
n+
n
N
N
the second set is represented by 2N1
0
..
.
0
, ...,
1
(3.40)
with an odd number of n

i s equal to 1.
These two sets belong to two distinct orbits of SO(2N )C+F . To see this one must study
the way they transform under
SO(2N

)C+F . Consider for instance the case of N = 2: the
3 0
+

+
exchange (n+
SO(4)C+F transformations 0 3 and 01 1
1 , n2 ) (n1 , n2 ) and (n1 , n1 )
0
(n+
2 , n2 ), respectively. In the general SO(2N ) case, two solutions differing by the exchange
+ +
+
+
(ni , nj ) (n
i , nj ) or (ni , ni ) (nj , nj ) for some i, j , therefore belong to the same orbit
of SO(2N)C+F . The vortices in the set (3.39) belong to a continuously degenerate set of minimal
vortices; the set (3.40) form the basis of another, degenerate set. The two sets do not mix under
the SO(2N ) transformations.
In order to see better what these two sets might represent, and to see how each vortex
transforms under SO(2N )C+F , let us assign the two states, |j , |j of a j th ( 12 ) spin,
N
j = 1, 2, . . . , N , to the pair of vortex winding numbers (n+
j , nj ) = (0, 1), (1, 0). Each of the 2
N
minimum vortices (Eqs. (3.39), (3.40)) can then be represented by the 2 spin state,
|s1 |s2 |sN ,
|sj = | = (0, 1),
or
| = (1, 0).
For instance the first vortex of Eq. (3.39) corresponds to the state, | . . . .
(3.41)
97
Introduce now the gamma matrices as direct products of N Pauli matrices acting as
j 3 3 1 1 1

(j = 1, 2, . . . , N );
(3.42)
j 1
N+j 3 3 2 1 1

(j = 1, 2, . . . , N).
(3.43)
j 1
k , k = 1, 2, . . . , 2N , satisfy the Clifford algebra

{i , j } = 2ij ,
i, j = 1, 2, . . . , 2N,
and the SO(2N ) generators can accordingly be constructed by ij = 4i1 [i , j ]. SO(2N ) transformations (including finite transformations) among the vortex solutions can thus be represented
by the transformations among the N -spin states, (3.41).
As each of ij (i = j ) flips exactly two spins, the two sets (3.39) and (3.40) clearly belong to
two distinct orbits of SO(2N ). In fact, a chirality operator
5 P
2N

j ,
{5 , j } = 0,
j = 1, 2, . . . , 2N,
(3.44)
j =1
anticommutes with all j s, where P = 1 (N even) or P = i (N odd), hence commutes with

SO(2N). The two sets Eq. (3.39), Eq. (3.40) of minimal vortices thus are seen to transform as
two spinor representations of definite chirality, 1 and 1, respectively (with multiplicity 2N1
each).
Every minimal solution is invariant under a U (N ) group embedded in SO(2N )C+F . This can
be seen from the form of the first solution in (3.39) in the basis (3.36):
2
1
..
..
q(1) = f+ (r, )
(3.45)
+ f (r, )
.
.
.
1
This solution is invariant under the subgroup U (N ) SO(2N ) acting as U q(1) U T , where U
U (N ) commutes with the second matrix in (3.45).
In the N -spin state representation above, the vortex (3.45) corresponds to the state with all
spins down, | . . . . In order to see how the N -spin states transform under SU(N ) SO(2N ),
construct the creation and annihilation operators
1
aj = (j iN+j ),
2
1
aj = (j + iN+j ),
2
satisfying the algebra,

!
{aj , ak } = aj , ak = 0,
!
aj , ak = j k .
SU(N ) generators acting on the spinor representation, can be constructed as [27]

Ta =
a j t a j k ak ,
j,k
98
where t a are the standard N N SU(N ) generators in the fundamental representation. The state
| . . . is clearly annihilated by all T a , as it is annihilated by all
ak = 3 3 1 1,

k = 1, 2, . . . N,
k1
thus, the vortex (3.45) leaves U (N ) invariant.

All other solutions can be obtained as Rq(1) R T with R O(2N ), so each solution is invariant
under an appropriate U (N ) subgroup RU R T . This means that the moduli space contains two
copies of the coset space
M = SO(2N )/U (N ).
(3.46)
The points in each coset space transform according to a spinor representation of definite chirality,
each with dimension 2N1 . When discussing the topological properties of vortices, we will see
that these disconnected parts correspond to different elements of the homotopy group.
Vortices of higher windings are described by N0 > 1. In the simplest non-minimal case, the
vortices are described by:
2 0
2 0
2 0
2 0
1 1
2 0 2 0 2 0
1 1 1 1
. . . . . .

.. .. , .. .. , .. .. , . . . , ... ... , ... ... .
N0 = 2,
(3.47)

2 0 2 0 2 0
1 1 1 1
2 0
0 2
1 1
1 1
1 1
These orbits correspond to parts of the moduli space whose structure corresponds to the coset
spaces SO(2N)U (N k)SO(2k), where k is the number of (1, 1) pairs. Analogously vortices
with N0 3 can be constructed.
The argument that the minimum vortices transform as two spinor representations implies that
the N0 = 2 vortices (3.47) transform as various irreducible antisymmetric tensor representations
of SO(2N)C+F , appearing in the decomposition of products of two spinor representations: e.g.,
2N1 2N1 or 2N1 2N1 .
(3.48)
Although all these vortices are degenerate in the semi-classical approximation, non-BPS corrections will lift the degeneracy, leaving only the degeneracy among the vortices transforming as
an irreducible multiplet of the group SO(2N )C+F . For instance the last vortex n+
a = na = 1, for
all a, carries only the unit U (1) winding and is a singlet, the second last vortex and analogous
ones belong to a 2N, and so on.
Due to the fact that the tension depends only on N0 = 2n(0) (twice the U (1) winding) the
degeneracy pattern of the vortices does not simply reflect the homotopy map which relates the
vortices to the massive monopoles. The monopole-vortex correspondence will be discussed in
Section 5 below.
The profile functions (f0 , fa , a+ , a ) for the simplest non-minimal vortex, N0 = 2 are illustrated in Fig. 3. In the figure is just considered the two simplest elements (n+ , n ) = (1, 1)
and (n+ , n ) = (2, 0). Adding elements of the same type corresponds just to a rescaling of the
2 and of the functions as in the minimal vortex case (N = 1). Adding elements
coupling g2N
0
a
of different types ((2, 0) or (1, 1)) does not induce new behavior.
99
Fig. 3. Numerically integrated minimum vortex solution with N0 = 2, where we have taken the couplings to be
2 = 1. In the left panel we have shown the element (n+ , n ) = (1, 1) and in the right panel the ele4g12 = 4g2N
The dependence of the couplings turns out to be similar to the case of the minimal vortex
ment (n+ , n ) = (2, 0).
(n+ , n ) = (1, 0). (i 2fi ).
4. Vortices in SO(2N + 1) theories

Consider now the case of a theory with symmetry breaking
v1
v2
SO(2N + 3) SO(2N + 1) U (1) 1.
(4.1)
The fields which remain massless after the first symmetry breaking can be found exactly as in the
even SO theories by use of various SO(3) groups, leading to Eq. (2.15), with A = 1, 2, . . . , Nf
where we now take Nf = 2N + 1. The light quarks can get color-flavor locked VEVs as in
Eq. (3.11), leading to a vacuum with global SO(2N + 1)C+F symmetry.
The ansatz (3.36) must be modified as follows
0
0
M1 (r, )
.
.
.
..
..
..
..
.
q(r, ) =
(4.2)
,
.
..
MN (r, )
0
(r)
0
ei n
introducing a new integer n and a new profile function (r).
Eq. (3.30) becomes

+ 2 2
df0
+ + ||
2 ,
= 2 2g12 r
a
a
dr
a
while the condition of finite energy gives

()
=
,
2N + 1
h0 () N0
,
=
n =
2
2
and the equation for (r)
is

h0
f0
d
= n
= .
dr
r
2
2r
(4.3)
(4.4)
(4.5)
(4.6)
100
(a)
(b)
Fig. 4. Numerically integrated minimum vortex solution of the SO(2N +1) theory, with N0 = 2 and we take the couplings
2 = 1. In the left panel we have (n+ , n ) = (1, 1) and in the right panel (n+ , n ) = (2, 0). The
to be 4g12 = 4g2N
1
1
1
1
dependence of the couplings turns out to be analogous to the case of the (n+
1 , n1 ) = (1, 0) vortex. (i 2fi ).
Note that the condition (4.5) fixes n in terms of N0 : as n must be an integer, this theory
contains only vortices with even N0 . This can be traced to the different structure of the gauge
groups. In fact, SO(2N + 3) has no center, so the pattern of symmetry breaking is
SO(2N + 3) SO(2N + 1) U (1) 1,
(4.7)
and there are no vortices with half-integer winding around the U (1), or around any other Cartan
U (1) subgroups.
The vortices are classified by the same integers n
a as before, but now there are SO(2N +
1)C+F transformations which exchange n+

a na singly. The minimal vortices are labeled by
2 0
2 0
2 0
1 1
2 0 2 0
1 1 1 1
. . . .
+ . . . .
.
.
. . . .
.
.
na , na = . . , . . , . . . ,
(4.8)
n = 1.
. . , . . ,
2 0 2 0
1 1 1 1
2 0
1 1
1 1
1 1
The moduli space contains subspaces corresponding to these orbits, whose structure is that of the
coset spaces SO(2N + 1)(U (N k) SO(2k + 1)) where k is the number of (1, 1) pairs.
The vortex profile functions are shown in Fig. 4.
5. Monopoles, vortices, topology and confinement
5.1. Homotopy map
The multiplicity of vortex solutions depends on the particular topology of the symmetrybreaking pattern of our model.
Usually, in systems with a gauge Lie group G and a symmetry-breaking pattern
v1
v2
G H 1,
there are:
(5.1)
101
Stable Dirac monopoles, classified by 1 (G);

Regular monopoles, classified by 2 (G/H ); topologically stable only in the limit v2 0;
Vortices, classified by 1 (H ); if they correspond to a non-trivial element of 1 (G), they are
topologically stable; otherwise they are topologically stable only in the limit v1 .
Monopoles and vortices are related by the topological correspondence [8]
2 (G/H ) = 1 (H )/1 (G),
(5.2)
so regular monopoles correspond to vortices which are trivial with respect to 1 (G), while vortices which are non-trivial with respect to 1 (G) correspond to Dirac monopoles.
In our theories of type DN , however, the center CG = Z2 acts trivially on all fields and the
breaking pattern is
v1
v2
G H CG ,
(5.3)
and the topological relation (5.2) is not directly useful. In fact, vortices are classified by
1 (H /CG ), which is a richer homotopy group than 1 (H ) 2 (G/H ) 1 (G). In our example
the relevant group is

SO(2N) U (1)
1
(5.4)
= Z Z2 .
Z2
The failure of (5.2) would mean that the correspondence between monopoles and vortices is lost.
Actually, it is better to formulate the problem as follows. The theory contains only fields in
the adjoint representation, so we can neglect the center CG from the beginning and consider the
gauge group as G = G/CG . In our example, the gauge group of the high-energy theory can
be taken as G = SO(2N + 2)/Z2 , broken to H = (SO(2N ) U (1))/Z2 at scale v1 and then
completely broken at scale v2 :
v1
v2
G H 1.
(5.5)
instead of Eq. (5.3). Then the relation (5.2) reads

(1)
1 SO(2NZ)U
SO(2N + 2)
2
=
2

.
SO(2N ) U (1)
1 SO(2N+2)
Z2
(5.6)
Regular monopoles are classified by the same homotopy group as before, because
SO(2N + 2)/Z2
SO(2N + 2)
=
,
(SO(2N) U (1))/Z2 SO(2N ) U (1)
(5.7)
while for Dirac monopoles the situation is different: the relevant homotopy group is not
1 (SO(2N + 2)), but the larger group 1 (SO(2N + 2)/Z2 ) (see Appendix B)

SO(4J )
1
(5.8)
= Z2 Z2 ,
Z2
while
SO(4J + 2)
Z2

= Z4 ,
so that the Dirac monopoles have quantized Z2 Z2 or Z4 charges.
(5.9)
102
This means that the theory has a larger set of monopoles, and the correspondence between
monopoles and vortices (which confine them) is rather subtle.2
In Appendix B we briefly review the structure of the homotopy groups which are relevant for
this analysis.
Finally, for the groups of type BN , the situation is slightly simpler as there is no non-trivial
center. The non-trival element of 1 (SO(2N + 3)) = Z2 represents the (unique type of) Dirac
monopoles; the elements of 1 (SO(2N + 1) U (1)) = Z2 Z label the vortices of the lowenergy theory. The vortices whose (non-trivial) winding in the group SO(2N + 1) U (1)
corresponds to a contractible loop in the parent theory, confine the regular monopoles.
5.2. Flux matching
To establish the matching between regular GNO monopoles and low-energy vortices, we use
the topological correspondence discussed in the previous section. Dirac monopoles are classified by 1 (SO(2N + 2)/Z2 ) or by 1 (SO(2N + 3)) depending on the gauge group, but regular
SO(2N+2)
SO(2N+3)
monopoles are classified by 2 ( SO(2N
)U (1) ) or by 2 ( SO(2N+1)U (1) ), i.e., homotopically nontrivial paths in the low-energy gauge group, which are trivial in the high-energy gauge group.
Regular monopoles can be sources for the vortices corresponding to these paths.
The
" vortices of the lowest tension which satisfy this requirement are those with N0 = 2
and a (n+
a na )/2 odd, so vortices corresponding to minimal GNO monopoles belong to the
SO(2N)C+F orbits classified by (3.47) with an odd number of (2, 0) pairs.
For a better understanding of this correspondence, we can also use flux matching between
vortices and monopoles [22]. There are 2N GNO monopoles obtained by different embeddings
of broken SU(2) SO(4) in SO(2N + 2). In a gauge where is constant, their fluxes are

(5.10)
d S B a t a = 2 2(t0 ti ),
S2
where t0 ti is the unbroken generator of the broken SU(2) subgroup. In the same gauge, the
flux of a vortex is

(5.11)
d 2 x Bza t a = N0 2t0 + n+
2tj ,
j nj
R2
so the fluxes agree for N0 = 2, n+

j nj = 2ij . The antimonopoles correspond to the opposite sign N0 = 2.
5.3. Monopole confinement: The SO(2N ) theory

We have now all the tools needed to analyze the duality in the SO theories at hand. The
general scheme for mapping the monopoles and vortices has been set up in Section 5.1. An
important point to keep in mind is that, while the vortex tension depends only on the U (1) flux
in our particular model (Eq. (3.26)), the classification of vortices according to the first homotopy
2 Note that the Lagrangian and fields for the two theories with gauge group SO(2N + 2) and SO(2N + 2)/Z are
2
the same. The set of vortices is the same for both theories and has a topological correspondence with the larger set of
monopoles.
103
group reflects the other Cartan charges (windings in SO(2N ) or SO(2N + 1)). It is necessary
to keep track of these to see how the vortices in the low-energy theory are associated with the
monopoles of the high-energy system.
First consider the theories of type DN , with the symmetry breaking
v1
v2
SO(2N + 2) SO(2N ) U (1) 1,
(5.12)
studied in detail in the preceding sections. The vortices with minimum winding, N0 = 1, of
Eqs. (3.39), (3.40), correspond to the minimum non-trivial element of 1 ((SO(2N ) U (1))/Z2 ),
which represent also the minimal elements of 1 (SO(2N + 2)/Z2 ). This last fact means that they
are stable in the full theory. They would confine Dirac monopoles of the minimum charge in the
underlying theory, 1 of Z4 or (1, 0) or (0, 1) of Z2 Z2 , see Appendix B.2.
Consider now the vortices Eq. (3.47) with N0 = 2. As the fundamental group of the underlying
theory is given by either Eq. (5.8) or Eq. (5.9), some of the vortices will correspond to noncontractible loops in the underlying gauge group: they would be related to the Dirac monopoles
and not to the regular monopoles. Indeed, consider the last of Eq. (3.47):

1 1 1 ... 1
na
=
.
(5.13)
n+
1 1 1 ... 1
a
It is characterized by the windings n(0) = 1, n(a) = 0 for all a. Thus it is an ANO vortex of the
U (1) theory, with no flux in the SO(2N ) part. It corresponds to a 2 rotation in (12) plane in
the original SO(2N + 2) groupthe path P in Appendix B.1: it is to be associated with a Dirac
monopole of charge 2.
The vortices of the type

0 1 1 ... 1
,
(5.14)
2 1 1 ... 1
and analogous ones (with (2, 0) or (0, 2) appearing in different positions) are characterized by
the two U (1) windings only: a flux n(0) = 1 and one of the Cartan flux of SO(2N ), e.g., n(1) = 1
(n(a) = 0, a = 1). They correspond to a simultaneous 2 rotations in (12) and in (34) planes
in the gauge group and it represents a contractible loop in the high-energy gauge group. They
confine regular monopoles, as can be seen also by the flux matching argument discussed in
Section 5.2.
Part of the continuous moduli of these vortex solutions include
SO(2N )U (1) SO(2N 2),
(5.15)
as the individual soliton breaks SO(2N )C+F symmetry of the system. This space corresponds to
the complex quadric surface Q2N 2 (C). As these vortices are not elementary but composite of
the minimal vortices, determining their correct moduli space structure is not a simple task.
Nevertheless, there are some indications that these correspond to a vector representation 2N
of SO(2N)C+F , appearing in the decomposition of the product of two spinor representations,
Eq. (3.48). In fact, the vortex Eq. (5.14) arises as a product

0 0 0 ... 0
0 1 1 ... 1
,
(5.16)
1 1 1 ... 1
1 0 0 ... 0
i.e., a product of two spinors of the same chirality if N is odd; vice versa, of spinors of opposite chirality if N is even. This corresponds precisely to the known decomposition rules in
SO(4m + 2) and SO(4m) groups (see, e.g., [27], Eq. (23.40)).
104
In order to establish that these vortices indeed transform under the SO(2N )C+F as a 2N one
needs to construct the moduli matrix [24] for these, and study explicitly how the points in the
moduli space transform. This problem will be studied elsewhere.
It is interesting to note that there seems to be a relation between the transformation proper#
ties of monopoles under the dual GNO group SO(2N
) and the transformation properties of the
corresponding vortices under the SO(2N )C+F group. In fact, vortices transforming as a vector
#
), as
of SO(2N)C+F have precisely the net magnetic flux of regular monopoles in 2N of SO(2N
classified by the GNO criterion.
Other vortices in Eq. (3.47) correspond to various Dirac (singular) or regular monopoles in
different representations of SO(2N )C+F .
5.4. Monopole confinement: The SO(2N + 1) theory
In the BN theories with the symmetry breaking
v1
v2
SO(2N + 3) SO(2N + 1) U (1) 1,
(5.17)
the minimal vortices of the low-energy theory have N0 = 2. Reflecting the difference of 1 group
of the underlying theory as compared to the DN cases (Z2 as compared to Z2 Z2 or Z4 ), the
N0 = 1 vortices (with half winding in U (1) and SO(2N )) are absent here.
The minimal vortices (4.8) again correspond to different homotopic types and to various
SO(2N + 1) representations. The vortex

1 1 1 ... 1
,
n = 1,
(5.18)
1 1 1 ... 1
has the U (1) charge n(0) = 1 and no charge with respect to SO(2N + 1). It is associated to the
non-trivial element of 1 (SO(2N + 3)) = Z2 : it is stable in the full theory. Its flux would match
that of a Dirac monopole. This is a singlet of SO(2N + 1)C+F (its moduli space consists of a
point).
Consider instead the vortices

0 1 1 ... 1
,
n = 1,
(5.19)
2 1 1 ... 1
and analogous ones, having the winding numbers n(0) = 1, n(a) = 1, n(b) = 0, b = a, and
n = 1. These would correspond to regular monopoles which, according to GNO classification,
are supposed to belong to a 2N representation of the dual group USp(2N ). Again, though it
is not a trivial task to establish that these vortices do transform as 2N of such a group, there
are some hints they indeed do so. It is crucial that the symmetry group (broken by individual
soliton vortices) is SO(2N + 1): it is in fact possible to identify the 2N generators constructed
out of those of SO(2N + 1), that transform them appropriately (Appendices A, B). Secondly,
the flux matching argument of Section 5.2 do connect these vortices to the minimum, regular
monopoles appearing in the semiclassical analysis. As in the DN theories these observations
should be considered at best as a modest hint that dual group structure as suggested by the
monopole-vortex correspondence is consistent with the GNO conjecture.
6. Conclusions
In this paper we have explicitly constructed BPS, non-Abelian vortices of a class of SO(N )
U (1) gauge theories in the Higgs phase. The models considered here can be regarded as the
105
bosonic part of softly broken N = 2 gauge theories with Nf quark matter fields. The vortices
considered here represent non-trivial generalizations of the non-Abelian vortices in U (N ) models
widely studied in recent literature.
The systems are constructed so that they arise as low-energy approximations to theories in
which gauge symmetry suffers from a hierarchical breaking
v1
v2
SO(N + 2) SO(N ) U (1) 1,
v1 v 2 ,
(6.1)
leaving an exact, unbroken global (SO(N ) U (1))C+F symmetry. Even though the low-energy
SO(N ) U (1) model with symmetry breaking
v2
SO(N ) U (1) 1,
(6.2)
can be studied on its own right, without ever referring to the high-energy SO(N + 2) theory,
consideration of the system with hierarchical symmetry breaking is interesting as it forces us to
try (and hopefully allows us) to understand the properties of the non-Abelian monopoles in the
v1
high-energy approximate system with SO(N + 2) SO(N ) U (1) and their confinement by
the vorticeslanguage adequate in the dual variablesfrom the properties of the vortices via
homotopy map and symmetry argument. Note that in this argument, the fact that the monopoles
in the high-energy theory and the vortices in the low-energy theory are both almost BPS but not
exactly so, is of fundamental importance [9,28].
In the models based on SU(N ) gauge symmetry, the efforts along this line of thought seem to
be starting to give fruits, giving some hints on the nature of non-Abelian duality and confinement.
Although the results of this paper are a only a small step toward a better and systematic understanding of these questions in a more general class of gauge systems, they provide a concrete
starting point for further studies.
Acknowledgements
This work is based on a master thesis by one of us (L.F.) [26]. The authors acknowledge useful
discussions with Minoru Eto, Muneto Nitta, Giampiero Paffuti and Walter Vinci. L.F thanks also
Roberto Auzzi, Stefano Bolognesi, Jarah Evslin and Giacomo Marmorini for useful discussions
and advices.
Appendix A. SO(2N), USp(2N), SO(2N + 1)
The change of basis to the one where a vector multiplet 2N of SO(2N ) naturally breaks to
under U (N ), is given by (see Eq. (2.28))
N+N
3
Q
q3
..
..
.
.
q2N+1
2
i1/
2
1/
2N+1
(A.1)
iQ = i1/ 2
q4 .
1/ 2
3
..
..
.
.
q2N+2
iQ2N+1
The SO(2N) generators,

E
F
,
tF D
(A.2)
106
where D, E, F are all pure imaginary N N matrices, with the constraints t E = E, t D = D,

are accordingly transformed as

1/ 2 i/ 2
E
F
1/ 2 i/ 2
tF D
i/ 2 1/ 2
i/ 2 1/ 2

t
1
(E + D) + i(F + F ) i(E D) + (F t F )
.
=
(A.3)
2 i(E D) + (F t F ) (E + D) i(F + t F )
Since both E, D are antisymmetric, (E + D) in the 1st block is the most general antisymmetric
imaginary matrix, while i(F + t F ) is the most general symmetric real matrix. Their sum gives
the most general N N hermitian matrix, which corresponds to generators of U (N ). In other
words, the subgroup U (N ) SO(2N ) is generated by those elements with E = D, F = t F .
On the other hand, the generators of USp(2N ) group have the form

B
A
,
(A.4)
C tB
with the constraints, t A = A, t C = C, A = C, B = B. The fact that A is symmetric while
the non-diagonal blocks in Eq. (A.3) are antisymmetric, means that there is no further overlap
between the two groups, that is, the maximal common subgroup between SO(2N ) and USp(2N )
is U (N ).
It is possible to get a hint on how USp(2N ) groups can appear as transformation group of
Q) under which the latter
the vortices. In order to see transformations among the vortices (Q,
could transform as 2N, it is necessary to embed the system in a larger group, such as SO(2N + 1)
model considered in Section 4. The idea is to build a map3 between the SO(2N + 1) generators (antisymmetric matrices) and the USp(2N ) generators which have the form, Eq. (A.4). The
ith SO(4) SU(2) SU(2) subgroup is generated by (with a simplified notation (1, 2, 3, 4)
(1, 2, 2i + 1, 2i + 2))
i
i
i
T2 = (31 42 ),
T3 = (12 43 ). (A.5)
T1 = (23 41 ),
2
2
2
The two vortices living in this SO(4) group are taken to be ith and (N + i)th components of the
fundamental representation of USp(2N ). The pairs can be transformed to each other by rotations
in the (2i + 2, 2N + 3) plane ( SO(2N + 1)), thus
Ai,i = i2i+2,2N+3 .
(A.6)
T
On the other hand, the two vortices associated with subgroups

living in the (1, 2, 2i + 1,
2i + 2) subspace and those living in the (1, 2, 2j + 1, 2j + 2) subspace, j = i, are transformed
into each other by rotations in the (2i +1, 2i +2, 2j +1, 2j +2) space: they transform in SO(2N )
(in the subspace i = 3, 4, . . . , 2N + 2). We have already seen that they actually do transform as a
pair of U (N ) representations, in the basis Eq. (A.1). As the U (N ) elements are generated by the
SO(2N) infinitesimal transformations with E = D, F = t F , one finds the map,
Bi,j = i(2i,2j + 2i+1,2j +1 ) + (2i,2j +1 2i+1,2j ).
(A.7)
Non-diagonal elements Aij , i = j , can be generated by commuting the actions of (A.6) and (A.7).
3 This correspondence can be applied equally well to the minimal regular monopoles constructed semiclassically, and
has been discussed in this context in [25].
107
Appendix B. Fundamental groups

Let us briefly discuss the (first) homotopy groups relevant to us:
B.1. SO(2N + 2)
There is only one non-trivial closed path P in this case, the rotation from 0 to 2 around any
axis. The rotation from 0 to 4 is homotopically equivalent to the trivial path, so P 2 = 1 and the
homotopy group is

1 SO(2N + 2) = Z2 .
(B.1)
B.2. SO(2N + 2)/Z2
Actually, in the model discussed in this paper, all the fields are in the adjoint representation
of SO(2N + 2): the gauge group effectively corresponds to SO(2N + 2) modulo identification
1 = 1. The path P is again non-trivial, but now there are also two inequivalent closed paths P+
and P going from 1 to 1, defined as P+ P1 = P . Explicitly, they can be taken as simultaneous rotations in N + 1 planes

P+ : ei12 12
(B.2)
eii,i+1 i,i+1 , 12 : 0 , i,i+1 : 0 ,
i=3,5,...,N 1
P :
i12 12
eii,i+1 i,i+1 ,
12 : 0 ,
i,i+1 : 0 .
(B.3)
i=3,5,...,N 1
When N + 1 is even, P+2 = P2 = 1 and P+ P = P . The homotopy group is generated by

P+ , P :

SO(4N )
= Z2 Z2 .
1
(B.4)
Z2
When N + 1 is odd, P+2 = P2 = P and P+ P = 1, so the homotopy group is generated by P+
only, and is of cyclic order four

SO(4N + 2)
1
(B.5)
= Z4 .
Z2
B.3. (SO(2N ) U (1))/Z2
After the symmetry breaking at the higher mass scale v1 , the theory reduces to an (SO(2N )
U (1))/Z2 theory. The division by Z2 corresponds to the identification (1, 1) = (1, 1), inherited from the underlying theory. From the point of view of the low-energy effective theory, it is
due to the fact that all the light matter fields qA,j , qA,j are in the vector representation of SO(2N )
but they carry at the same time the unit charge with respect to U (1).
The non-trivial paths of SO(2N ) U (1) are combinations of Q (a 2 rotation in any plane in
SO(2N)) and the paths Rn winding n times around the U (1). The simplest non-trivial closed
paths that arise after the Z2 quotient are P+, 1 , P+, 1 , P, 1 , P, 1 going from (1, 1) to
2
2
2
2
(1, 1) with a half winding around U (1). By taking U (1) to act in the (12) plane, SO(2N )
in the (34 . . . N) space, they can be explicitly chosen as simultaneous rotations in (12), (34),
108
(56), . . . planes

ei12 12 ei34 34
eiii+1 i,i+1 ;
(B.6)
i=5,7,...,N1
with
P+, 1 :
2
12 : 0 ,
34 : 0 ,
P+, 1 : 12 : 0 ,
34 : 0 ,
P, 1 :
2
12 : 0 ,
i,i+1 : 0 ,
i,i+1 : 0 ,
34 : 0 ,
P, 1 : 12 : 0 ,
i,i+1 : 0 ,
34 : 0 ,
i,i+1 : 0 .
(B.7)
(B.8)
(B.9)
(B.10)
Note that P+, 1 and P+, 1 correspond respectively to the P+ and P paths in the SO(2N + 2)
2
2
theory.
When N is even, P+,a P+,b = P,a P,b = Ra+b and P+,a P,b = QRa+b , so every group
element can be written as (P+,1/2 )k Q with k Z, = {0, 1}. The homotopy group is

1
SO(2N) U (1)
Z2

= Z Z2 ,
N even.
(B.11)
When N is odd, P+,a P+,b = P,a P,b = QRa+b and P+,a P,b = Ra+b , and every group element can again be written as (P+,1/2 )k Q with k Z, = {0, 1}, as in the N even case. The
homotopy group is

SO(2N) U (1)
1
(B.12)
= Z Z2 , N odd.
Z2
Even though the homotopy group is the same for the two cases (N even or odd), its embedding
in 1 (SO(2N) U (1)) = Z Z2 is different: Rn corresponds to k = 2n, = 0 for N even and
to k = 2n, = 1 for N odd. In other words
R1 = (P+,1/2 )2 Q
(N odd);
R1 = (P+,1/2 )2
(N even).
(B.13)
B.4. Relation between the smallest elements of the high-energy and low-energy fundamental
groups
There are simple relations among the smallest elements of the groups 1 ( SO(2N+2)
) and
Z2
(1)
). From the above explicit constructions one sees that
1 ( SO(2NZ)U
2
P+ = P+, 1 ;
P = P+, 1 = R1 P+, 1 ;
(B.14)
and by using Eq. (B.13), one has

$
P+, 1 =
2
(P+, 1 )1 Q,
odd N,
(P+, 1
even N.
2
2
)1 ,
(B.15)
109
B.5. SO(2N + 3)
The fundamental group is Z2 as in the SO(2N + 2) cases, and the smallest closed path being
P:
eiij ij : ij = 0 2,
in any plane (ij ). P 2 = 1 and the homotopy group is

1 SO(2N + 3) = Z2 .
(B.16)
(B.17)
B.6. SO(2N + 1) U (1)

At the mass scales below v1 the theory reduces to an SO(2N + 1) U (1) theory with matter in the fundamental representation, q and q carrying charges 1 with respect to U (1). The
fundamental group is

1 SO(2N + 1) U (1) = Z2 Z,
(B.18)
where Z represents the number of winding (charge) in the U (1) part and Z2 a 2 rotation in any
plane in SO(2N + 1).
References
[1] E. Lubkin, Ann. Phys. 23 (1963) 233;
E. Corrigan, D.I. Olive, D.B. Fairlie, J. Nuyts, Nucl. Phys. B 106 (1976) 475.
[2] P. Goddard, J. Nuyts, D. Olive, Nucl. Phys. B 125 (1977) 1.
[3] F.A. Bais, Phys. Rev. D 18 (1978) 1206.
[4] E.J. Weinberg, Nucl. Phys. B 167 (1980) 500;
E.J. Weinberg, Nucl. Phys. B 203 (1982) 445;
K. Lee, E.J. Weinberg, P. Yi, Phys. Rev. D 54 (1996) 6351.
[5] S. Coleman, The Magnetic Monopole Fifty Years Later, Lectures given at International School of Subnuclear
Physics, Erice, Italy, 1981.
[6] A. Abouelsaood, Nucl. Phys. B 226 (1983) 309;
P. Nelson, A. Manohar, Phys. Rev. Lett. 50 (1983) 943;
A. Balachandran, G. Marmo, M. Mukunda, J. Nilsson, E. Sudarshan, F. Zaccaria, Phys. Rev. Lett. 50 (1983) 1553;
P. Nelson, S. Coleman, Nucl. Phys. B 227 (1984) 1.
[7] N. Dorey, C. Fraser, T.J. Hollowood, M.A.C. Kneipp, Non-Abelian duality in N = 4 supersymmetric gauge theories,
hep-th/9512116;
N. Dorey, C. Fraser, T.J. Hollowood, M.A.C. Kneipp, Phys. Lett. B 383 (1996) 422, hep-th/9605069.
[8] R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi, H. Murayama, Nucl. Phys. B 701 (2004) 207, hep-th/0405070.
[9] M. Eto, L. Ferretti, K. Konishi, G. Marmorini, M. Nitta, K. Ohashi, W. Vinci, N. Yokoi, Nucl. Phys. B 780 (3)
(2007) 161, hep-th/0611313.
[10] K. Konishi, Magnetic Monopole Seventy-Five Years Later, Lecture Notes in Physics, vol. 737, Springer, 2007,
p. 471, hep-th/0702102, in honor of the 65th birthday of Gabriele Veneziano.
[11] P.C. Argyres, M.R. Plesser, N. Seiberg, Nucl. Phys. B 471 (1996) 159;
P.C. Argyres, M.R. Plesser, A.D. Shapere, Nucl. Phys. B 483 (1997) 172;
K. Hori, H. Ooguri, Y. Oz, Adv. Theor. Math. Phys. 1 (1998) 1.
[12] A. Hanany, Y. Oz, Nucl. Phys. B 466 (1996) 85.
[13] G. Carlino, K. Konishi, H. Murayama, JHEP 0002 (2000) 004;
G. Carlino, K. Konishi, H. Murayama, Nucl. Phys. B 590 (2000) 37;
G. Carlino, K. Konishi, S.P. Kumar, H. Murayama, Nucl. Phys. B 608 (2001) 51.
[14] S. Bolognesi, K. Konishi, Nucl. Phys. B 645 (2002) 337;
S. Bolognesi, K. Konishi, G. Marmorini, Nucl. Phys. B 718 (2005) 134.
[15] G. Marmorini, K. Konishi, N. Yokoi, Nucl. Phys. B 741 (2006) 180.
110
[16] K. Hashimoto, D. Tong, JCAP 0509 (2005) 004, hep-th/0506022.

[17] R. Auzzi, M. Shifman, A. Yung, Phys. Rev. D 73 (2006) 105012, hep-th/0511150.
[18] M. Eto, K. Konishi, G. Marmorini, M. Nitta, K. Ohashi, W. Vinci, N. Yokoi, Phys. Rev. D 74 (2006) 065021,
hep-th/0607070.
[19] A. Hanany, D. Tong, JHEP 0307 (2003) 037;
A. Hanany, D. Tong, JHEP 0404 (2004) 066.
[20] R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi, A. Yung, Nucl. Phys. B 673 (2003) 187.
[21] M. Shifman, A. Yung, Phys. Rev. D 70 (2004) 045004;
A. Gorsky, M. Shifman, A. Yung, Phys. Rev. D 71 (2005) 045010.
[22] R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi, Nucl. Phys. B 686 (2004) 119, hep-th/0312233.
[23] D. Tong, TASI lectures on solitons: Instantons, monopoles, vortices and kinks, hep-th/0509216.
[24] Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, Phys. Rev. D 71 (2005) 065018;
M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, J. Phys. A 39 (2006) R315, hep-th/0602170.
[25] L. Ferretti, K. Konishi, Duality and confinement in SO(N ) gauge theories, FestSchrift, Sense of Beauty in Physics,
in honor of the 70th birthday of A. Di Giacomo, Edizioni PLUS University of Pisa Press, 2006, hep-th/0602252.
[26] L. Ferretti, Vortici non abeliani e gruppi duali in teorie di gauge N = 2SO(N ) e USp(2N ), master thesis at University
of Pisa, 2004, unpublished.
[27] H. Georgi, Lie Algebras in Particle Physics, second ed., Westview Press, 1999.
[28] M.J. Strassler, JHEP 9809 (1998) 017, hep-th/9709081;
M.J. Strassler, On Phases of Gauge Theories and the Role of Non-BPS Solitons in Field Theory, III Workshop,
Continuous Advance in QCD, University of Minnesota, 1998, hep-th/9808073.
[29] M. Eto, K. Hashimoto, G. Marmorini, M. Nitta, K. Ohashi, W. Vinci, Phys. Rev. Lett. 98 (2007) 091602, hep-th/
0609214.
Can we study quark matter in the quenched

approximation?
Pietro Giudice a , Simon Hands b,
a Dipartimento di Fisica Teorica, Universit di Torino and INFN, Sezione di Torino,
via P. Giuria 1, I-10125 Torino, Italy

b Department of Physics, Swansea University, Singleton Park, Swansea SA2 8PP, UK
Received 8 March 2007; received in revised form 29 June 2007; accepted 25 July 2007
Available online 7 August 2007
Abstract
We study a quenched SU(2) lattice gauge theory in which, in an attempt to distinguish between timelike
and spacelike gauge fields, the gauge ensemble {U } is generated from a 3-dimensional gauge-Higgs model,
the timelike link variables being reconstructed from the Higgs fields. The resulting ensemble is used to
study quenched quark propagation with non-zero chemical potential ; in particular, the quark density,
chiral and superfluid condensates, meson, baryon and gauge-fixed quark propagators are all studied as
functions of . While it proves possible to alter the strength of the inter-quark interaction by changing the
parameters of the dimensionally reduced model, there is no evidence for any region of parameter space
where quarks exhibit deconfined behaviour or thermodynamic observables scale as if there was a Fermi
surface.
PACS: 11.15.Ha; 12.38.Gc; 12.38.Mh
Keywords: Quenched approximation; Non-zero chemical potential
1. Introduction
Lattice QCD at non-zero quark chemical potential ought in principle to be much more
straightforward than the corresponding problem with non-zero temperature T . The reason is
that can be introduced via a local term in the Lagrangian, whereas T > 0 is imposed via
E-mail address: s.hands@swan.ac.uk (S. Hands).

doi:10.1016/j.nuclphysb.2007.07.025
112
P. Giudice, S. Hands / Nuclear Physics B 789 (2008) 111132
a lattice of finite extent Lt a = T 1 in the temporal direction. Simulations with varying can
therefore be performed at fixed lattice spacing a, so that the renormalisation factors needed for
bulk thermodynamic observables such as the Karsch coefficients required to extract the physical
energy density need be calculated only once, rather than for each value of T studied. Moreover,
the regime /T 1 relevant for the physics of the superdense matter found in neutron star cores
implies lattices with large Lt , so that temporal correlators can be sampled efficiently without
recourse to anisotropic lattices requiring careful calibration. This means that excitations above
the ground state of the system, or more generally the nature of the spectral density function, can
be studied with relative ease. Indeed, this programme has been successfully carried out in certain
theories such as the NJL and related models, where both measurement of the superfluid gap in
the quasiparticle spectrum at k [1] and the identification of a phonon excitation with k
in the spin-1 meson channel [2] have proved possible.
In practise, of course, the reason this happy state of affairs has not been exploited in QCD
is the notorious Sign Problem associated with most Euclidean field theories having a non-zero
density of a conserved charge. In brief, for lattice QCD with Nf quark flavors described by
a Lagrangian density of the form qMq,
the functional measure detNf M() = detNf M ()

implying that for = 0 the action is complex, rendering Monte Carlo importance sampling
impracticable in the thermodynamic limit. Can one then at least perform lattice QCD simulations
in the quenched Nf 0 limit, i.e., study the propagation of valence quarks with = 0 through
a non-perturbative gluon background?
In principle the information extracted from such an approach could be at best qualitative,
since (unlike the case of T > 0) the gauge field ensemble {U } can only respond to = 0 via
virtual quark loops, so that in an orthodox quenched simulation the gluon background is that of
the vacuum with zero baryon charge density. Nonetheless, such information might be valuable,
for instance, in furnishing a non-perturbative definition of the Fermi surface, whose existence is
assumed in most phenomenological treatments of dense matter.
In the NJL studies referred to above, the Fermi surface appears as a minimum of the dispersion relation (k) in the neighbourhood of the Fermi momentum k = kF : for weakly-interacting
massless quarks we expect kF . Even free quark propagation at = 0 reveals the presence
of a Fermi surface and so is not entirely trivial. For QCD however, the notion of a distinguished
quark momentum is not gauge invariant, so that the identification of regions of phase space occupied by low-energy excitations may be altogether more subtle in a gauge invariant formalism.
However, as plausible as this sounds, the quenched approach has conceptual difficulties at
= 0. In the context of a random matrix theory, Stephanov [3] showed that the quenched theory
should be thought of as the Nf 0 limit of a QCD-like theory with not just Nf flavors of quark
As has
q 3 of the SU(3) gauge group but also with Nf flavors of conjugate quark q c 3.
been known for many years [4], the extra particle content results in gauge invariant qq c bound
states in the spectrum which in a strongly interacting theory can result in baryons degenerate
with light mesons. At = 0 these extra states are usually regarded as extra pions due to the
fermion species doubling resulting from the use of the manifestly real positive measure det MM
required by practical fermion algorithms. The introduction of = 0 distinguishes baryons from
mesons. A simple argument, which assumes that binding energy is a relatively small component
of the energy density of bulk nuclear matter, predicts that for /T 1 there should be an onset
transition from the vacuum to a ground state with quark number density nq > 0 once equals
the mass of the lightest baryon divided by the number of quark constituents. For QCD this scale
is mN /3, where mN is the nucleon mass; for a theory with conjugate quarks the onset scale is
m /2. Only cancellations among configurations due to a complex-valued measure det M can
113
ensure that the fake signal for nq > 0 vanishes in the range m /2 < < mN /3 [5]. A recent
analytic demonstration has been given, once again in the context of a random matrix model,
in [6].
It is clear from these considerations that quenched calculations at = 0 can only be useful
if the distribution of gluon fields is modified in some way to reflect high baryon density. For
instance, a minimum requirement is that the timelike links U0 are no longer drawn from the
same sampling distribution as spacelike links Ui , reflecting the breakdown of Lorentz invariance
due to the preferred rest frame of the background quark distribution. If the gluons were modified
in some way so that color confinement no longer holds, then the role of qq c excitations may not
be so important in determining the ground state in the quark sector, and it is at least conceivable
that valence quark propagation in such a background may qualitatively resemble that of the
deconfined regime of the phase diagram at /T 1 corresponding to quark matter. The focus of
such a study would thus be cold quarks in hot glue. While there are several lines one could take,
the approach we shall explore in this study is to start with the 3d configurations characteristic of
the deconfined phase found at T > Tc , /T 1 produced by the approach to hot gauge theory
known as Dimensional Reduction (to be reviewed below in Sections 2.1 and 2.2). In this approach
all non-static modes of the gauge theory (i.e., those with non-zero Matsubara frequency) are
integrated out leaving a 3d gauge-Higgs model describing the non-perturbative behaviour of the
remaining static modes. For sufficiently large T /Tc the model coefficients are perturbatively
calculable functions of T , and Nf , and the resulting effective theory can be used to make
quantitative predictions in the quarkgluon plasma phase. Our goals are less ambitious and more
speculative; we will use the Higgs fields of the 3d theory to reconstruct the timelike gauge
fields U0 and hence generate (3 + 1)d configurations of static gauge fields suitable for the study
of valence quark propagation with = 0. At this stage we are simply trying to generate a nonconfining gluon background, and in no sense claim to be developing a high density effective
theory.
With gauge group SU(3) and = 0, the dimensional reduction (DR) machinery yields a
cubic Higgs self-interaction with imaginary coefficient proportional to Nf [7], which is how the
3d effective theory inherits the sign problem from (3 + 1)d QCD. Whilst there is no reason to
suppose the sampling of the models configuration space would be any less problematic than that
of the full theory, at least the complex phase of each configuration is now expressed in terms of
a local term in the action, making its calculation cheap. Moreover, since the coefficient of the
complex term can be made arbitrarily small ad hoc, the effect of gradually introducing the sign
problem, and the interplay of the complex phase with physical observables, could be explored.
In this paper, however, we focus on the technically simpler case of gauge group SU(2). In this
case the action is real for even Nf (and the cubic DR Higgs self-interaction vanishes), permitting
orthodox lattice simulations with standard algorithms, e.g. [8].
QCD with gauge group SU(2), often referred to as Two Color QCD or QC2 D, has been studied
at = 0 with lattice simulations by several groups using a variety of formulations and algorithms
[5,812]. Since q and q fall in equivalent representations of the gauge group, so that in effect q
and q c are identical, hadron multiplets contain both q q mesons and qq, q q baryons, which are
degenerate at = 0. In the chiral limit the lightest hadrons are Goldstone bosons, and can be
analysed using chiral perturbation theory ( PT) [13]. At leading order for T a second order
onset transition from vacuum to matter consisting of tightly bound diquark scalar bosons is pre starts to fall below its
dicted at exactly o = m /2. At the same point the chiral condensate
qq
vacuum value, and a non-vanishing diquark condensate
qq develops, such that
qq
2 +
qq2
remains constant. The diquark condensate spontaneously breaks U(1) baryon number symmetry,
114
so the resulting ground state is superfluid. In the limit o+ the matter in the ground state
becomes arbitrarily dilute, weakly-interacting, and non-relativistic, and is a textbook example
of BoseEinstein condensation. This scenario was subsequently confirmed by simulations with
staggered lattice fermions [5,9]. More recent simulations have found evidence for a second transition at larger to a deconfined phase, as evidenced by a non-vanishing Polyakov loop [11]
and by a fall in the topological susceptibility [12]. In this regime thermodynamic quantities scale
according to the expectations of free field theory (also referred to as StefanBoltzmann (SB)
scaling), namely nq 3 , and energy density 4 [11].
In this paper we build on our existing experience by exploring = 0 in quenched QC2 D.
In Section 2 below we specify our procedure for generating (3 + 1)d quenched SU(2) gauge
configurations starting from a 3d gauge-Higgs model, and review standard quark observables
once = 0. Our results follow in Section 3. First we explore the quark density nq and the chiral
qq
and superfluid
qq condensates to see how the equation of state responds to attempts to
render the (3 + 1)d theory non-confining, and whether the behaviour predicted by PT can be
supplanted by the SB scaling expected of weakly-interacting degenerate quarks. Next we turn to
spectroscopy, calculating both normal and anomalous components of the quark propagator,
taking advantage of the enormous gain in statistical accuracy offered by the quenched approach.
First we study the bound state spectrum in both meson and diquark sectors, finding evidence
for significant mixing between the sectors at large enough . Next, for the first time in a gauge
theory, we present results for the quark propagator at = 0 obtained following gauge-fixing.
A range of spatial momenta are sampled in an attempt to map out the quasiquark dispersion relation. While we find significant qualitative differences between strongly and weakly-interacting
quarks, no signal for a Fermi surface has emerged. Our conclusions follow in Section 4.
2. Formulation
2.1. Reconstructing the fourth dimension
The quenched action we start from is the 3d SU(2) gauge-adjoint Higgs model given by
Eq. (4) of Ref. [14]:

1
tr(x x )
1 tr Ux,ij + 2
S3d =
2
x
x,i>j

2

+
2 tr(x x ) 1 ,
tr x Ux,i x+ Ux,i
2
(1)
x,i
is a traceless hermitian 2 2 matrix representing the adjoint Higgs field. As is

where
usual in a gauge-Higgs model, increasing at sufficiently large takes one from a confinement
phase with small
tr to a Higgs phase with large
tr .
The action (1) is derived by dimensional reduction from a 4d pure gauge SU(2) theory with
non-zero temperature T . In continuum notation the original Lagrangian density L4 ( A )2
in terms of a gluon field with canonical mass dimension [A] = 1. The dimensionally reduced
Lagrangian is obtained by integrating over Euclidean time and discarding all non-static modes,
i.e., setting 0 = 0:
1
2 a a
T
L3 =
dx0 L4
0
1
1
(Di Aj )2 + (Di A0 )2 (Di Bj )2 + (Di )2 ,
T
T
(2)
115
with the new fields defined by Bi = T 1/2 Ai , = T 1/2 A0 , [B] = [] = 1/2. When transcribing to the 3d lattice action (1) we use

a
1/2
Ui = exp(igaAi ) = exp igaT Bi ,
(3)
=
,
with [U ] = [] = 0. The 3d lattice parameters , and are all dimensionless, with given in
terms of the YangMills coupling by
=
4N
4
4
2 = 2 ,
2
ag T
g
ag3
(4)
where it is helpful to distinguish between the dimensionful coupling g3 of the 3d theory and the
dimensionless coupling g of the parent hot 4d theory, related via g32 = g 2 T ; in the final equality
we have assumed that the 4d theory is formulated on a lattice with N time spacings.
In the DR approach to the effective description of hot field theory, the parameters , and
are perturbatively calculable functions of g, T and quark chemical potential , as outlined in [7].
The attractiveness of this approach is that effects of, say Nf flavors of massless quark or a small
quark chemical potential can be incorporated in the coefficient calculation, the result always
being a 3d bosonic model which is relatively cheap to simulate. The DR provides a good effective
description when the non-static modes decouple; the authors of [7,14] claim that this is valid for
T 2Tc , 4T . The parameter choice employed in this study is discussed further below in
Section 2.2.
As discussed in Section 1, the quenched approximation has long been considered to be inapplicable to QCD with = 0, because it includes unphysical light baryons formed as bound states
of q 3 and q c 3 [3,4]. The reason these qq c states are light and hence distort the physics is
because at T = 0 quenched QCD configurations are in a confining phase, implying spontaneous
chiral symmetry breaking. The lightest qq c state is degenerate with the Goldstone pion. Our goal
is to study quark propagation through a non-confining quenched gluon background with = 0.
0 ,
Since chemical potential couples to quarks via the timelike component of the current
this is an inherently four-dimensional problem. In order to generate such a background we take
a 3d configuration generated by the DR simulation, motivated by the fact that it describes deconfining physics, and reconstruct the gauge field in the timelike direction following (3) via the
prescription (recall aA0 = aT 1/2 = (aT )1/2 ):

a a
= cos(g a a ) + i
sin(g a a ),
U0 = exp ig
(5)
N
a a

with g = . Spatial link variables Ui are taken to be time independent and identical to their
3d counterparts. The resulting model differs from DR in that it has a non-trivial electrostatic
potential (i.e., a spatially-varying U0 field), but more importantly differs from real physics with
T in that there are no non-static (i.e., 0 = 0) modes.1 It is not derivable from QCD in any
systematic way.
It is legitimate to ask whether excluding non-static modes from consideration is too severe
an approximation to be physically reasonable. It is possible to examine this issue for degenerate
quark matter interacting weakly via gluon exchange [16]. The loop integral in the self-consistent
1 Even the static modes in dense matter with a sharp Fermi surface may exhibit oscillatory behaviour, known as Friedel
oscillations [15], which are absent from the DR approach.
116
equation for the color-superconducting gap is dominated by small angle scattering between
quarks at antipodal points of the Fermi sphere, diverging as ln where q0 /, q0 being the
timelike momentum of the exchanged gluon. The divergence must be cut off by some physical
screening mechanism: for electric gluons this is Debye screening, resulting in a min g below
which the interaction is effectively point-like; for magnetic gluons the mechanism is Landau
1/3
damping resulting in min g 2/3 q0 /1/3 . In the static limit q0 0 magnetic gluons are thus
unscreened in perturbation theory, and the resulting gap equation yields exp(3 2 / 2g)
[17]. The effect of neglecting non-static modes such as electric gluons can be shown to affect
the pre-exponential factor, but not the scaling of the gap with the coupling strength. Neglect of
non-static modes thus seems defensible in a weak-coupling approach to quark matter; beyond
perturbation theory we have little to guide us.
A conceptual point worth stressing is that due to the different ways spacelike and timelike
links are treated, there is no simple relation between spatial and temporal lattice spacings. In
principle one could determine the relation empirically by comparison of correlators in different directions, and as /at calibrated by matching, say, to the ratio of the physical (and T , dependent) Debye screening mass to a hadron mass measured at = 0. We make no attempt
to follow this prescription in this exploratory study: all masses and energy scales are compared
with a reference scale chosen to be the pion mass at = 0, calculated in dimensionless units as
m at . This prevents us for the time being from a direct comparison with condensates such as the
physically observable quark density, expressed in dimensionless units as nq as3 . Note, however,
that the dimensionless ratio /T can be expressed without ambiguity using lattice variables as
N ,
where N is the temporal extent of the lattice and = at is the dimensionless lattice
chemical potential introduced in Eq. (16).
To illustrate both the benefits and potential pitfalls of this approach it is helpful to consider
rectangular Wilson loops. The spatial Wilson loop W (r1 , r2 ) clearly inherits the behaviour of
the 3d theory; it is known to decay with an area law [18]. Confinement in the (3 + 1)d theory,
however, is governed by the temporal loop W (r, t) given by
W (r, t) =

1
0; r, 0)U t (r )P (0,
t; r, t) U (0)
t.
tr P (0,
0
0
Nc
(6)
Here P is the path-ordered product of spacelike links connecting 0 to r; because the configuration
0; r, 0) = P (0,
t; r, t) P . Using (4), (5) we deduce
is static these can be written P (0,
W (r, t) =
1
0 i2gt
tr P ei2gt
P e r.
Nc
(7)
Now expand the exponentials as power series in g,

noting that (2)2 = a a 1 ||2 1, and
n
tr = 0 for odd n, implying that odd terms vanish. The lowest non-trivial term is O(g 2 ):
4g 2 t 2

g 2 t 2
tr P 0 P r .
|0 |2 + |r |2 +
(8)
2!
Nc
To proceed we make some simplifying approximations. For any = 0, r > 0, the second
ab ; in the strong coupling limit 0, strong link field
term of (8) is proportional to 0a rb Padj
fluctuations should make its expectation value very small, and hence we neglect all terms of
this form. Secondly, when integrating over fluctuations of the we assume translational invariance and neglect all non-Gaussian fluctuations (which should be valid as 0), so that
|0 |2p |r |2q
||2(p+q) . This also requires physical excitations of the 3d theory to be massive so that
0 r can be neglected. It is then straightforward to show that all r-dependence
117
2n /(2 (2n)!). Hence

drops out and the expectation value of the O(g 2n ) term is (1)n (2gt
||)

|| .
W (r, t) cos2 gt
(9)
As required, the temporal Wilson loop is positive, well behaved for gt||
1, and clearly does

not obey an area law. The reconstructed (3 + 1)d theory is thus non-confining. Note that the
neglected P P terms always contribute to the series coefficients with the opposite sign (due to
an extra factor of i 2 ), so that their inclusion would in effect make
W decay less rapidly. Viewed
as a function of Euclidean time
W decays with a negative curvature; this is inconsistent
with

a transfer matrix with positive definite spectrum, which implies decays of the form i ci eEi t .
Unsurprisingly, therefore, the (3 + 1)d model violates unitarity.
2.2. Parameter choice for the dimensionally reduced model
The DR procedure for SU(N ) gauge theory with Nf flavours of fermions has been studied
to two loops in [14]. The effective theory obtained is a three-dimensional SU(N ) adjoint Higgs
theory, with the scalars corresponding to the electric gauge potential A0 in the unreduced theory;
the case with chemical potential = 0 is dealt with for example in Ref. [7].
The continuum action of the SU(2) adjoint Higgs model is given by [14]

S=

2
1
2
d x
tr(Fij Fij ) + tr(Di Dj ) + m3 tr() + 3 tr() ,
2
3
(10)
where Fij = i Aj j Ai + ig3 [Ai , Aj ] and Di = i + ig3 [Ai , ]. The physical properties
of the theory are fixed by the two dimensionless ratios
x=
3
,
g32
y=
m23
g34
(11)
In the framework of DR these parameters are completely determined by g 2 and T of the original
four-dimensional gauge theory. Since g 2 is a running coupling its value is fixed by the renormalisation scale MS . Choosing the renormalisation scale as in [19] and expressing MS through
the critical temperature Tc = 1.23(11)MS as measured on the lattice [20], it is possible to show
[19] that for Nf = 0
g32 =
10.7668
T,
ln(8.3T /Tc )
x=
0.3636
,
ln(6.6T /Tc )
y(x) =
1
2
+
+ O(x).
2
9 x 4 2
(12)
With these equations, specifying T /Tc completely fixes the parameters x, y of the reduced
model. Note that the values of x and y corresponding to Tc are about xc 0.19 and yc 0.14.
The model exhibits two phases: a symmetric phase with full SU(2)-type confinement and a Higgs
phase with residual U(1)-type confinement [21].
Simulations of the three-dimensional SU(2) adjoint Higgs model showed that this theory has
a phase diagram with continuously connected Higgs and confinement phases, which are partially
separated by a line of first order phase transitions [19,22]. The discretized form of the action (10)
in terms of a rescaled lattice field is simply our action (1). The 3d lattice parameters , and
are all dimensionless. The parameters of the continuum and lattice theory are related up to two
118
Table 1
The lattice parameters for = 9.0
T /Tc
1
2
3
5
10
100
0.193
0.141
0.122
0.104
0.087
0.056
0.142
0.185
0.210
0.242
0.285
0.427
0.3637812
0.3620027
0.3612335
0.3603992
0.3594500
0.3569257
0.0028379
0.0020531
0.0017689
0.0015009
0.0012490
0.0007927
loops by a set of equations [23],

4
, = 2,
2
ag3

2
1
2x
5
y=
3
+
1+ x
8
4
4

1
3
2
+
ln
+ 0.09 + 8.7 + 11.6x ,
20x 10x
2
16 2
x=
(13)
where = 3.17591. Due to the theorys superrenormalisability, these perturbative relations are
exact in the continuum limit and, based on experience, are accurate for all > 6 [14]. In Table 1 we report some sample lattice parameter sets obtained by calculating x and y according to
Eqs. (12) and fixing the ratio T /Tc ; then we have calculated and using Eqs. (13).
For gauge group SU(3) the DR theory is very similar to Eq. (10) [7]:

2
1
S = d 3x
(14)
tr Fij2 + tr[Di , A0 ]2 + m23 tr A20 + 3 tr A20
,
2
where Fij = i Aj j Ai + ig[Ai , Aj ], Di = i + ig3 Ai , Fij , Ai , and A0 are all traceless 3 3
Hermitian matrices (A0 = Aa0 Ta , etc.), and g32 and 3 are the gauge and scalar coupling constants,
with mass dimension one. The physical properties of the effective theory, also in this case, are
determined by the dimensionless ratios
x=
3
,
g32
y=
m23 ( 3 = g32 )
g34
(15)
where 3 is the MS dimensional regularisation scale in 3d.

Here we note that application of DR to QCD with Nf quark flavors and non-zero chemiN
cal potential results in a complex term ig 3 3f2 tr A30 [7], so that the DR theory inherits the
complex action of the parent theory (for gauge group SU(2) tr A30 0 and the action is real). In
principle a term of this form could be incorporated in our approach, the advantage being that a
Sign Problem could be introduced with arbitrarily small coefficient, and the response of observables to small phase fluctuations of the measure assessed. This gradualist approach is of
course impossible in full QCD simulations.
In applications of the DR action (1) to hot gauge theory, the parameter is used to control the
physical lattice spacing in units of Tc1 , and and are tuned according to relations (12), (13) to
specify T . A complication [14] is that for the {, , } parameter set appropriate for hot QCD the
3d model has its true ground state in the Higgs phase, whereas the perturbative continuation to
119
QCD (12) requires it to be in the confining phase.2 Fortunately the confining phase appears
to be metastable, so that simulations started off with small
tr can be used to yield physically
relevant results. Once the model is extended to 4d using (5) the physical meaning of the lattice
parameters is no longer clear. In this exploratory study we hold , fixed and, with the exception
of Figs. 9 and 10, restrict our attention to two values of .
2.3. Introducing quarks with = 0
Henceforth we use 4d configurations {U } generated as outlined above as input in quenched
studies using staggered fermions in the fundamental representation of SU(2), having action S1 +
S2 with
S1 =

1
0
x e 0 Ux, x+ e
Ux , x + m
x x M
2 x
x
(16)
and
S2 =
j
x
j
xtr 2 x + x 2 xtr .
2
(17)
As explained above, the lattice parameter is related to the physical quark chemical potential
via = at . Henceforth we will ignore the distinction between as , at when quoting values for
parameters such as quark mass m and diquark source j ; in consequence no attempt will be made
to compare these quantities with a physical scale.
The simplest observables to discuss are the chiral condensate and quark density:

1
1
G 1
G 1
=
(18)
tr
G ,
nq =
tr
G ,
2V
m
4V

where
denotes a quenched average using the action (1), and G is the Gorkov propagator
defined below in (25). The normalisation, somewhat arbitrary in a quenched simulation, is chosen
so that in the limit , nq saturates at a value of two per lattice site. The term S2 comprising
gauge invariant diquark source terms3 is introduced in order to discuss the possibility of diquark
condensation. The Pauli matrices 2 act on SU(2) color indices. While S1 is invariant under the
i , S2 is not. Defining source strengths j = j j,
global U(1) rotation ei , e
we obtain diquark condensates

1
G 1
qq =
(19)
G .
tr
2V
j
If we choose sources such that j+ = j+ = j , j = 0, then the condensate forms in the +
channel, and a massless Goldstone pole develops in the channel as j 0 as confirmed by
2 In this context the terms confining and Higgs refer to 3d dynamics, corresponding to the behaviour of spatial
Wilson loops in a (3 + 1)d model. They have no bearing on the behaviour either of the hot gauge theory before DR or the
reconstructed 4d model studied here.
3 Note that a gauge invariant diquark source does not exist for gauge group SU(3). Physically, this means diquark
condensation forces superfluidity for SU(2), but superconductivity for SU(3). Technically, it means diquark condensation
is more difficult to address by lattice methods for SU(3).
120
the Ward identity

qq+
qq (0)qq (x) =
.
j+
x
(20)
The numerical implementation of all these observables is identical to that for the NJL model
described in [1].
3. Numerical results
We have chosen = 9.0 sufficiently close to the continuum limit for the DR formalism to be
trustworthy, and start with a point with T = 2Tc according to (12); from Table 1 this corresponds
to = 0.3620027, = 0.0020531. For this parameter set, we have explored system volumes
83 Lt , with Lt = 16, 32, 64 and 128. For the most part in this section we present data taken
with Lt = 32. However, it is important to make a precise determination of the pion mass at = 0.
Fig. 1 shows data for the pion mass m as a function of diquark source j taken at = 0. The
data are extrapolated to j = 0 using a PT-inspired form [13] m (j ) = m (0)(1 + bj 2 )1/4 (the
bare quark mass ma = 0.05, so that the pion remains massive as j 0). We conclude m at (j =
0) = 0.230(3), consistent with a measurement made exactly at = j = 0: m at = 0.2321(1).
3
Since this scale is not too dissimilar to L1
s , we have repeated the measurement on 16 64,
where we find m at = 0.2368(3). The systematic error due to finite volume is significant, but
small enough at 2% to be acceptable for this exploratory study.
3.1. Equation of state
as functions of for
In Figs. 2 and 3 we plot quark density nq and chiral condensate
qq
various j , this time on a 83 32 lattice, with 0 /T 8. The bare quark mass throughout this
study was set to ma = 0.05. To confirm freedom from finite volume effects we also performed
test simulations with Ls = 4, 8, 16, 32 and Lt = 16, 32, 64. There is a transition at 0.12,
constant to one in which nq
becoming more abrupt as j 0, from a phase with nq = 0,
qq
increases approximately linearly with and
qq
2 . This is in complete accordance with the
scenario described by PT in which as increases at T 0 there is a transition at c = m /2
Fig. 1. m vs. j from data taken on 83 128 at ma = 0.05, = 0 for = 9.0, = 0.362007, = 0.0020531.
121
Fig. 2. nq vs. for various j using the same {, , }.
Fig. 3.
qq
vs. for various j using the same {, , }.
from the vacuum to a weakly-interacting Bose gas formed from scalar diquarks (cf. Figs. 4 and 5
of Ref. [13]). The diquarks are supposed to Bose-condense to form a superfluid condensate; on
a finite system this must be checked at j = 0 using the
qq+ observable of (19). Fig. 4 shows a
compilation of
qq+ data as a function of j for rising from zero up to = 0.25, the condensate
increasing monotonically with . To determine the nature of the ground state an extrapolation
j 0 is needed. We have used a cubic polynomial for data with 0.02 j a 0.1, which may
result in some systematic uncertainty in the immediate neighbourhood of the transition, but Fig. 5
confirms that once again there is an abrupt change of behaviour in the order parameter at
m /2, and that the high- phase is superfluid.
Next we explored a parameter set corresponding to a smaller scalar stiffness by changing
to = 0.1 while keeping ma = 0.05naively following Table 1 suggests this corresponds to a
huge value of T /Tc , i.e., taking us further into the deconfined phase of the hot gauge theory. Of
course, whether DR-based concepts remain valid for the reconstructed theory must be addressed
empirically. This time we used a volume 83 64 for the bulk observables and at = 0 determined
the pion mass m at = 0.1377(1) on 83 128, and m at = 0.1423(4) on 163 64, showing that
122
Fig. 4.
qq+ vs. j for various using the same {, , }.
Fig. 5.
qq+ vs. for various j using the same {, , }.
the finite volume error is now roughly 3%. Figs. 6 and 7 show respectively nq and
qq
as
functions of for 0 /T 12 in the same format as previously. It is noteworthy that for
> m /2 nq () is numerically very similar to the values found at = 0.3620027, whereas
for < m /2 the chiral condensate
qq
is significantly smaller, indicative of a weaker quark
anti-quark binding at this smaller . As before, there is a clear discontinuity in the observables
behaviour at c m /2, and the general picture is qualitatively very similar, suggesting that the
PT scenario is still applicable. Diquark binding is now also much weaker, however, as shown
by the
qq+ data of Fig. 8. Note that the source values j are greater than those of Fig. 5, but
the corresponding
qq+ are smaller in magnitude. Since the curvature of the data as a function
of j is quite pronounced, a reliable extrapolation j 0 is impracticable on this system size.
Indeed, inspection by eye would suggest a linear extrapolation would yield
qq+ 0 for all ;
however, the j a = 0.02 data do manifest some significant variation at m /2, suggestive that
weak symmetry-breaking persists for > c .
It is disappointing that we have found no qualitative change in physics as the parameters are
variedrecall that the PT model which describes the results reasonably well is based on the
Fig. 6. nq vs. for various j at = 9.0, = 0.1, = 0.0020531.
Fig. 7.
qq
vs. for various j using the same {, , }.
Fig. 8.
qq+ vs. for various j using the same {, , }.
123
124
Fig. 9. nq vs. for various ,

. Note nq increases systematically with .
Fig. 10. nq vs. for various , .
assumption of confinement, or at least on the presence of very tightly bound diquark states in the
spectrum. To explore the parameter space more widely we focussed on a single observable, nq ,
and scanned the (, ) plane on 83 16 at five different values of with = 9.0, ma = 0.05 and
j a = 0.01. The results are summarised in Figs. 9 and 10. Fig. 9 shows that except for = 0.1
the results for fixed are practically independent of (shown by the overlapping symbols) and
of (shown by the horizontal lines). Fig. 10 shows that nq increases linearly with over a wide
region of parameter space, as it does for > c in Figs. 2, 6. This approximate linear behaviour
is once again a prediction of PT [5,9,13], and is to be contrasted with the nq 3 behaviour
expected of a deconfined theory where baryons can be identified with degenerate quark states
occupying a Fermi sphere of radius kF . The absence of this scaling is a further reason to
conclude that the reconstructed model does not describe deconfined physics.
3.2. Bosonic spectrum
Next we report on bound state spectroscopy. Conventionally, q q bound states are referred to
as mesons, and qq, q q as diquark baryons and anti-baryons respectively. In a medium with spon-
125
taneously broken baryon number symmetry, however, excitations need not have a well-defined
baryon number. An appropriate set of states to look at, together with gauge-invariant interpolating operators expressed in terms of staggered fermion fields, is as follows:
pion
x x x ;
(21)
x x ;
(22)

1 t
2 x + x 2 xt qq+x ;
Higgs
(23)
2 x

1 t
x 2 x x 2 xt qqx .
Goldstone
(24)
2
Here the phase x = (1)x0 +x1 +x2 +x3 , and the Pauli matrix 2 acts on color indices. Pion and
scalar states are related via the U(1) global symmetry ei , ei . Analogous
to chiral symmetry for continuum spinors, this is an exact symmetry of the action (16) in the
limit m 0. In a phase with spontaneously broken chiral symmetry, the scalar is massive, and
the pion a Goldstone mode, becoming massless as m 0. Similarly, Higgs and Goldstone
i , an exact
diquark states are related via the U(1)B baryon number rotation ei , e
symmetry of (16), (17) in the limit j 0. In a superfluid phase with
qq+ = 0, the Higgs is
massive, and the Goldstone massless in the limit j 0.
The boson correlators are constructed from the Gorkov propagator [1]

j2
Axy Nxy
M() 1
=
Gxy =
(25)
,
j 2 xy
M tr ()
N xy A xy
scalar
where the 2 2 (in color space) matrices N

x y and A
x y are known as the normal
and anomalous parts respectively. On a finite volume A 0 for j = 0; limj 0 limV A = 0
signals particlehole mixing resulting from the breakdown of U(1)B symmetry, and hence superfluidity. Due to SU(2) symmetries the only independent components of G are Re N11 N
and Im A12 A and their barred counterparts, so that in practise all information from a single
source can be extracted at = 0 using just two matrix inversions per configuration. Each of the
correlators C(t) constructed from the forms (21)(24) receives contributions from diagrams containing both N and A-type propagators; by construction, however, they remain symmetric under
t t even once = 0. Note also that in this quenched treatment contributions to the mesons
from disconnected N loops and to the diquarks from disconnected A loops are neglected.
Fig. 11 shows all four timeslice correlators evaluated on a 83 128 lattice with parameter set
= 9.0, = 0.3620027 and = 0.0020531. The chemical potential = 0.25, i.e., above the
critical c required to enter the superfluid phase. The quark bare mass ma = 0.05 and diquark
source j a = ja = 0.02. All four channels yield clear signals for single particle bound states, the
Higgs being the noisiest.
Like any meson constructed from staggered fermions, the correlators in principle describe two
states and must be fitted using the form

C(t) = A emt + em(Lt t) + B eMt + (1)t eM(Lt t) ,
(26)
where m and M denote the masses of states with opposite parities. Fits to (26), where positive,
are shown by solid lines in Fig. 11. In most cases we find M m; however for > c the pion
correlator has a distinct saw-tooth shape, and in fact the fit yields m > Mb1 , where denotes
the usual pseudoscalar pion, and b1 a state of opposite parity, which must therefore be scalar
(note that the logarithmic scale requires Fig. 11 to plot |C (t)|).
126
Fig. 11. Timeslice propagators of various bosonic bound states at = 0.25 on a 83 128 lattice at = 9.0,
= 0.3620027, = 0.0020531, ma = 0.05, j a = 0.02.
Fig. 12. Mass spectrum of the pion and its parity partner as a function of .
Smaller symbols denote data taken on
163 64.
In Fig. 12 we plot m and Mb1 against , and in Fig. 13 the corresponding spectrum for all
the states (21)(24). Fig. 12 also shows m and Mb1 as measured on a 163 64 system at some
1
( = 0.9) 10 < Ls ;
representative values of . Even for the lightest measured mass, Mb1
while volume effects are statistically significant, they have no impact on the qualitative trends
we now discuss. First note that all states are approximately degenerate at = 0. The equality of
pion, Higgs and Goldstone correlators is guaranteed by SU(2) symmetry at = 0 [8], but the
degeneracy of the scalar in the chirally-broken vacuum suggested by Fig. 3 can only arise as a
result of mesondiquark mixing due to j = 0. Next, note that the pion mass remains constant
for < c , where as reviewed above it is a pseudo-Goldstone boson associated with chiral
symmetry breaking, and then falls once the superfluid phase is entered. This is in accordance with
the predictions of PT for the so-called PS state of a theory with Dyson index D = 4 [13],
and has also been observed in simulations with dynamical staggered quarks in the fundamental
representation of SU(2) [9]. Most of the other states, including the b1 in Fig. 12, show a much
steeper decrease with for < c , followed by a gentle rise to a plateau at ma 0.13 in the
superfluid phase > c , precisely that expected of the Goldstone state expected in the superfluid
127
Fig. 13. Mass spectrum of various bosonic excitations as a function of .
phase with diquark source j = 0 (cf. the QI state shown in Fig. 3 of [13]). The exception is
the Higgs, which rises more steeply to become the heaviest state at large . We conclude (a) the
breaking of degeneracy between Higgs (23) and Goldstone (24) states is clear supplementary
evidence for the breaking of U(1)B symmetry in the superfluid phase; (b) all states with J P = 0+
including the b1 but except the Higgs have some projection onto the Goldstone state, regardless
of whether the original interpolating operator is mesonic or baryonic. It would be interesting to
study this phenomenon as j is varied.
3.3. Fermionic spectrum
We have also studied the fermion spectrum, often in the context of condensed matter called
the quasiparticle spectrum. Since the Gorkov propagator G is not gauge invariant we have to
specify a gauge fixing procedure. A feature of the quenched approach is that it permits large
statistics to be accumulated with relatively little CPU effort. This has enabled us for the first time
in a gauge theory context to study G at = 0, by helping to overcome the sampling problems
associated with gauge fixing. We have experimented with two gauge choices: Unitary gauge
= (0, 0, 3 ), which is implemented before the reconstruction of the 4th dimension, and
is unique up to a Z2 factor, specified by demanding 3 0; and Coulomb gauge, implemented by

), in an attempt to make the gauge fields as smooth as possible

maximising xi tr(Ux,i + Ux
,i
and hence improve the signal-to-noise ratio.
In Figs. 14 and 15 we plot respectively the normal and anomalous fermion timeslice propagators on a 32 82 64 lattice at = 9.0, = 0.1, = 0.0020531, ma = 0.05, j a = 0.0.2
and = 0.3, the last value chosen to ensure > c . These plots result from an analysis of 800
independent configurations. As discussed in Ref. [1], the properties of staggered lattice fermions
are such that in a phase with approximate chiral symmetry, as expected for > c from Fig. 7,
the numerically important contributions to N (t) are from t odd, and to A(t) from t even, and
only these points are plotted. Most of the data was obtained using Coulomb gauge and differing
values of the momentum kx a {0, 16

, . . . , 2 }, though the kx = 0 data taken in unitary gauge
are shown for comparison. Two features to note are that the Coulomb data is roughly twice as
large as the unitary data reflecting an enhanced signal as anticipated above, and that there is little
variation with kx .
128
Fig. 14. Normal component of the quark propagator N (t) for various momenta k for > c .
Fig. 15. Anomalous component of the quark propagator A(t) for various momenta k for > c .
In the NJL model the quasiparticle propagator can be successfully fitted using the forms [1]
N (t) = P eEN t + QeEN (Lt t) ,

A(t) = R eEA t eEA (Lt t) ,
(27)
(28)
where for = 0 there is no reason to expect P = Q, but for a well-defined quasiparticle state
the equality EN = EA should hold. Note that the anomalous amplitude R = 0 is evidence for
baryon number symmetry breaking, often in a condensed matter context called particlehole
mixing. Fits to the data of Figs. 14 and 15 are shown in Table 2, and it should be noted that
only the anomalous channel fits produced an acceptable 2 . Some insight can be gained from
Fig. 16, which compares the Coulomb gauge normal and anomalous propagators at kx = 0 on
a logarithmic scale; while the fit (28) for A(t) looks plausible, the normal channel never settles
to a well-defined quasiparticle pole, even with Lt = 64. Moreover only the anomalous channel
shows any evidence of gauge independence in Table 2. The value of EA obtained is very close to
m /2, indicating that at this value of the pion is a weakly bound state. Another striking feature
of the data is the approximate forwardsbackwards symmetry of N (t), implying P Q.
129
Table 2
Fitted mass values to the data of Figs. 14 and 15 for kx = 0
EN
EA
Coulomb gauge
2 /dof
Unitary gauge
2 /dof
0.1244(7)
0.0674(13)
202
3.4
0.1651(17)
0.0696(15)
182
5.9
Fig. 16. Comparison of the kx = 0 data of Figs. 14 and 15 on a logarithmic scale.
Fig. 17. Dispersion relations EN (k) and EA (k).
Fig. 17 shows the dispersion relations EA (kx ) for data taken on a 32 82 64 lattice at
the parameter values shown (we were unable to obtain satisfactory fits to (27) to extract EN
for kx = 0). It confirms that the quasiparticle excitation energies are k-independent. This should
be contrasted with the findings of [1], where a lattice study of the NJL model using identical
formalism found E(k) exhibiting a pronounced minimum at k kF , the Fermi momentum,
evidence that the NJL model has a well-defined Fermi surface with kF . One motivation for
our study was to investigate to what extent the concept of a Fermi surface, which is not strictly
gauge invariant, can be put on a firm empirical footing in a gauge theory. Fig. 17 shows no
evidence for a Fermi surface.
The nature of the quasiparticle excitation is clarified a little at the other parameter set studied,
namely = 0.3620027 (with all other parameters unchanged). In this case our results show
130
Fig. 18. Close-up of A(t) for = 0.3620027 for various .
Fig. 19. E and vs. for = 0.3620027 in both normal and anomalous channels.
no evidence for any well-defined spin-1/2 state in either normal or anomalous channels; as a
result of confinement the excitation spectrum of the model seems to be saturated by the tightly
bound spin-0 states of Fig. 13. Fig. 18 shows a close up of A(t) for various values, showing
the presence of an oscillatory component whose amplitude initially grows with (the = 0.3
points overlay those from = 0.2), but whose wavelength is roughly -independent. The origin
of the oscillation could possibly be associated with the non-unitarity of the model discussed in
Section 2.1, but is most likely a manifestation of independent spin-1/2 excitations being illdefined due to confinement. We have fitted the = 0.3620027 data to the forms

N (t) = P eEN t cos(N t + ) + QeEN (Lt t) cos N (Lt t) + ,
(29)

E t
E
(L
t)
t
cos A (Lt t) + ,
A(t) = R e A cos(A t + ) e A
(30)
where we interpret E as the energy and as the width of a quasiparticle excitation. The results
are plotted in Fig. 19. Most of the results are taken in Coulomb gauge, but at = 0.3 data from
Fig. 20.
131

E 2 + 2 vs. for = 0.3620027 in both normal and anomalous channels.
unitary gauge is also available. The results most striking feature is their independence of ,
with of the same order of magnitude as E. An interesting systematic
effect is that EN > EA
while N < A , which has motivated us in Fig. 20 to plot E 2 + 2 vs. . The figure has
the same vertical scale as the previous two, and it is clear that the disparity between normal
and anomalous channels is significantly reduced. Inspection of the = 0.3data also shows
that the gauge dependence of this result is O(20%) at worst. Numerically, E 2 + 2 > m ,
indicating strong quarkanti-quark binding, due to the persistence of confinement at this value
of . Therefore we can interpret the effect of confinement as rotating the quasiparticle pole into
the complex plane, the rotation angle being larger in the anomalous channel than in the normal
one. It would be interesting if this feature could be reproduced by analytic methods such as
self-consistent solution of SchwingerDyson equations.
4. Conclusion
Our attempt to alter the nature of the gluon background by changing the parameters of the
3d DR gauge-Higgs model has been a partial success, in that in going from = 0.3620027 to
= 0.1 the strength of the binding between quarks weakens significantly. The main evidence
for this claim comes from the spectrum; at the smaller studied quasiquarks appear to be welldefined independent degrees of freedom, whose excitation energy is to fair precision half that of
the gauge-invariant pion state. At = 0.3620027 by contrast, the quasiquark propagator exhibits
a pole at complex k, and the resulting estimates for both energy and width of the excitation
exceed the pion mass. However, in neither case is there evidence for significant departure of nq ,
qq
and
qq from the behaviour predicted by PT, so that even if quarks are important degrees
of freedom at = 0.1, there is no evidence for the formation of a degenerate system signalled
by SB scaling. Moreover, our attempts to measure the quasiquark dispersion relation E(k) have
been unsuccessful; while there is evidence for the gauge-invariance of the minimum energy or
gap at = 0.1, no k-dependence is observed. This is, of course, entirely consistent with the
well-known fact that the only gauge invariant feature of a propagator is the location of its poles.
We conclude that identification of a Fermi surface, supposing one existed, presents a technical
challenge in any gauge theory simulation.
It is clear that our simplistic treatment, despite being manifestly gauge invariant and inclusive
of non-pointlike interactions, has missed some essential component of the physics of high density. If we had first attempted the quenching programme using gauge group SU(3), the departures
from the theoretical expectation of SB scaling in the large- limit could have been ascribed either
to unexpected non-perturbative effects, or to some subtle cancellation due to the Sign Problem
entirely missing from the quenched approach. For SU(2), however, orthodox simulations of full
QC2 D, in which the Sign Problem is absent, yield evidence for deconfinement and SB scaling
132
at large [11,12], which our approach misses completely. We can of course speculate on which
important features of the gluon background at high quark density are absent; possible candidates
2kF 2 [15]. Sadly though, it appears to remain
are non-static modes, and modes with |k|
the case that despite its unreasonable effectiveness in virtually all other aspects of lattice QCD,
the quenched approximation has nothing useful to tell us about the physics of high quark density.
Acknowledgements
We thank Owe Philipsen for generously giving us access to the gauge-Higgs simulation code.
The numerical work was performed on the PC cluster Majorana of the INFN Gruppo Collegato di Cosenza at the Universit della Calabria (Italy).
References
[1] S.J. Hands, D.N. Walters, Phys. Rev. D 69 (2004) 076011.
[2] S.J. Hands, J.B. Kogut, C.G. Strouthos, T.N. Tran, Phys. Rev. D 68 (2003) 016005;
S.J. Hands, C.G. Strouthos, Phys. Rev. D 70 (2004) 056006.
[3] M.A. Stephanov, Phys. Rev. Lett. 76 (1996) 4472.
[4] A. Gocksch, Phys. Rev. D 37 (1988) 1014;
A. Gocksch, Phys. Rev. Lett. 61 (1988) 2054.
[5] S.J. Hands, I. Montvay, S.E. Morrison, M. Oevers, L. Scorzato, J.I. Skullerud, Eur. Phys. J. C 17 (2000) 285;
S.J. Hands, I. Montvay, L. Scorzato, J.I. Skullerud, Eur. Phys. J. C 22 (2001) 451.
[6] J.C. Osborn, Phys. Rev. Lett. 93 (2004) 222001;
G. Akemann, J.C. Osborn, K. Splittorff, J.J.M. Verbaarschot, Nucl. Phys. B 712 (2005) 287.
[7] A. Hart, M. Laine, O. Philipsen, Nucl. Phys. B 586 (2000) 443.
[8] S.J. Hands, J.B. Kogut, M.-P. Lombardo, S.E. Morrison, Nucl. Phys. B 558 (1999) 327.
[9] J.B. Kogut, D.K. Sinclair, S.J. Hands, S.E. Morrison, Phys. Rev. D 64 (2001) 094505.
[10] R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Phys. Lett. B 493 (2000) 189;
S. Muroya, A. Nakamura, C. Nonaka, Phys. Lett. B 551 (2003) 305;
P. Giudice, A. Papa, Phys. Rev. D 69 (2004) 094509;
P. Cea, L. Cosmai, M. DElia, A. Papa, JHEP 0702 (2007) 066;
S. Chandrasekharan, F.J. Jiang, Phys. Rev. D 74 (2006) 014506.
[11] S.J. Hands, S. Kim, J.I. Skullerud, Eur. Phys. J. C 48 (2006) 193.
[12] B. Alles, M. DElia, M.P. Lombardo, Nucl. Phys. B 752 (2006) 124.
[13] J.B. Kogut, M.A. Stephanov, D. Toublan, J.J.M. Verbaarschot, A. Zhitnitsky, Nucl. Phys. B 582 (2000) 477.
[14] A. Hart, O. Philipsen, Nucl. Phys. B 572 (2000) 243.
[15] J. Kapusta, T. Toimela, Phys. Rev. D 37 (1988) 3731.
[16] See e.g., J.B. Kogut, M.A. Stephanov, The Phases of Quantum Chromodynamics, Cambridge Univ. Press, 2004
(Chapter 9.2).
[17] D.T. Son, Phys. Rev. D 59 (1999) 094019.
[18] C. Borgs, Nucl. Phys. B 261 (1985) 455;
C. DeTar, Phys. Rev. D 32 (1985) 276;
T.A. DeGrand, C.E. DeTar, Phys. Rev. D 34 (1986) 2469.
[19] K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 503 (1997) 357.
[20] J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493.
[21] S. Nadkarni, Nucl. Phys. B 334 (1990) 559.
[22] A. Hart, O. Philipsen, M. Teper, J. Stack, Phys. Lett. B 396 (1997) 217.
[23] M. Laine, Nucl. Phys. B 451 (1995) 484.
Stckelberg axions and the effective action

of anomalous Abelian models.
A SU(3)C SU(2)W U (1)Y U (1)B model
and its signature at the LHC
Claudio Corian a, , Nikos Irges b , Simone Morelli a
a Dipartimento di Fisica, Universit del Salento and INFN Sezione di Lecce, Via Arnesano, 73100 Lecce, Italy
b Department of Physics and Institute of Plasma Physics, University of Crete, 71003 Heraklion, Greece
Received 12 March 2007; received in revised form 29 June 2007; accepted 30 July 2007
Dedicated to the memory of Hidenaga Yamagishi
Abstract
We elaborate on an extension of the Standard Model with a gauge structure enlarged by a single anomalous U (1), where the presence of a WessZumino term is motivated by the GreenSchwarz mechanism
of string theory. The additional gauge interaction is anomalous and requires an axion for anomaly cancelation. The pseudoscalar implements the Stckelberg mechanism and undergoes mixing with the standard
Higgs sector to render the additional U (1) massive. We consider a 2-Higgs doublet model. We show that
the anomalous effective vertices involving neutral currents are potentially observable. We clarify their role
in the case of simple processes such as Z , which are at variance with respect to the Standard Model.
A brief discussion of the implications of these studies for the LHC is included.
1. Introduction
Among the possible extensions of the Standard Model (SM), those where the SU(3)C
SU(2)W U (1)Y gauge group is enlarged by a number of extra U (1) symmetries are quite
attractive for being modest enough departures from the SM so that they are computationally
E-mail address: claudio.coriano@le.infn.it (C. Corian).

doi:10.1016/j.nuclphysb.2007.07.027
134
C. Corian et al. / Nuclear Physics B 789 (2008) 133174
tractable, but at the same time predictive enough so that they are interesting and even perhaps
testable at the LHC. Of particular popularity among these have been models where at least one
of the extra U (1)s is anomalous, that is, some of the fermion triangle loops with gauge boson
external legs are non-vanishing. The existence of this possibility was noticed in the context of the
(compactified to four dimensions) heterotic superstring where the stability of the supersymmetric vacuum [1] can trigger in the four-dimensional low energy effective action a non-vanishing
FayetIliopoulos term proportional to the gravitational anomaly, i.e., proportional to the anomalous trace of the corresponding U (1). The mechanism was recognized to be the low energy
manifestation of the GreenSchwarz anomaly (GS) cancellation mechanism of string theory.1
Most of the consequent developments were concentrated around exploiting this idea in conjunction with supersymmetry and the FroggattNielsen mechanism [2] in order to explain the mass
hierarchies in the Yukawa sector of the SM [3], supersymmetry breaking [4], inflation [5] and axion physics [6], in all of which the presence of the anomalous U (1) is a crucial ingredient. In the
context of theories with extra dimensions the analysis of anomaly localization and of anomaly inflow has also been at the center of interesting developments [7,8]. The recent explosion of string
model building, in particular in the context of orientifold constructions and intersecting branes
[10,11] but also in the context of the heterotic string [12], have enhanced even more the interest
in anomalous U (1) models. There are a few universal characteristics that these vacua seem to
possess. One is the presence of U (1) gauge symmetries that do not appear in the SM [13,14]. In
realistic four-dimensional heterotic string vacua the SM gauge group comes as a subgroup of the
ten-dimensional SO(32) or E8 E8 symmetry [15], and in practice there is at least one anomalous U (1) factor that appears at low energies, tied to the SM sector in a particular way, which
we will summarize next. For simplicity and reasons of tractability we concentrate on the simplest
non-trivial case of a model with gauge group SU(3)C SU(2)W U (1)Y U (1)B where Y is
hypercharge and B is the anomalous gauge boson and with the fermion spectrum that of the SM.
The mass term for the anomalous U (1)B appears through a Stckelberg coupling [14,16,17] and
the cancellation of its anomalies is due to four-dimensional axionic and ChernSimons terms (in
the open string context see the recent works [14,1820]).
Despite of all this theoretical insight both from the top-down and bottom-up approaches, the
question that remains open is how to make concrete contact with experiment. However, as mentioned above, in models with anomalous U (1)s one should quite generally expect the presence
of a physical axion-like field and in fact in any decay that involves a non-vanishing fermion
triangle like the decay Z , Z , Z, Z Z , etc., one should be able to see traces of
the anomalous structure [19,20,22,23]. In this paper we will mostly concentrate on the gauge
boson decays which, even though hard to measure, contain clear differences with respect to the
SMas is the case of the Z decayand in addition with respect to anomaly free U (1)
extensionslike the Z decayfor example.
In [19] a theory which extends the SM with this minimal structure (for essentially an arbitrary
number of extra U (1) factors) was called Minimal Low Scale Orientifold Model or MLSOM
for short, because in orientifold constructions one typically finds multiple anomalous U (1)s.
Here, even though we discuss the case of a single anomalous U (1) which could also originate
from heterotic vacua or some field theory extension of the SM, we will keep on using the same
terminology keeping in mind that the results can apply to more general cases. We finally mention
1 Conventionally in this paper we will use both the term GreenSchwarz (GS) to denote the mechanism of cancelation
of the anomalies, to conform to the string context, though the term WessZumino (WZ) would probably be more
adequate and sufficient for our analysis. The corresponding counterterm will be denoted, GS or WZ, with no distinction.
135
that other similar constructions with emphasis on other phenomenological signatures of such
models have appeared before in [18,2426]. A perturbative study of the renormalization of these
types of models is in [27]. Other features of these models, in view of the recent activity connected
to the claimed PVLAS result [28], have been discussed in [23].
Our work is organized as follows. In the first sections we will specialize the analysis of [19] to
the case of an extension of the SM that contains one additional anomalous Abelian U (1), with an
Abelian structure of the form U (1)Y U (1)B , that we will analyze in depth. We will determine
the structure of the entire Lagrangean and fix the counterterms in the one-loop anomalous effective action which are necessary to restore the gauge invariance of the model at quantum level.
The analysis that we provide is the generalization of what is discussed in [23] that was devoted
primarily to the analysis of anomalous Abelian models and to the perturbative organization of
the corresponding effective action. After determining the axion Lagrangean and after discussing
Higgs-axion mixing in this extension of the SM, we will focus our attention on an analysis of
the contributions to a simple process (Z ). Our analysis, in this case, aims to provide an
example of how the new contributions included in the effective actionin the form of one-loop
counterterms that restore unitarity of the effective actionmodify the perturbative structure of
the process. A detailed phenomenological analysis is beyond the scope of this work, since it
requires, to be practically useful for searches at the LHC, a very accurate determination of the
QCD and electroweak background around the Z/Z resonance. We hope to return to a complete
analysis of 3-linear gauge interactions in this class of models in the near future.
2. Effective models at low energy: The SU(3)C SU(2)W U (1)Y U (1)B case
We start by briefly recalling the main features of the MLSOM starting from the expression of
the Lagrangean which is given by
1 G G 1 W W 1 B B 1 Y Y
Tr F F
F F
F
F F
L = Tr F
2
2
4
4

2
j
B

q
+ + ig2 Wj + igY quY AY + igB u B Hu
2
2

2
j

qdB
j
Y Y

+ + ig2 W + igY qd A + igB
B Hd
2
2

a
j
Li i + ig3 Ga + ig2 Wj + igY q (QL ) AY + igB q (QL ) B QLi
+Q
Y
B
2
2

(u )
(u )
+ u Ri i + igY qY R AY + igB qB R B uRi

(d )
(d )
+ dRi i + igY qY R AY + igB qB R B dRi

j j
(L) Y
(L)
+ Li i + ig2 W + igY qY A + igB qB B Li

2

(e )
(e )
+ eRi i + igY qY R AY + igB qB R B eRi

( )
( )
+ Ri i + igY qY R AY + igB qB R B Ri
L Hd dR u Q
L (i2 Hu )uR + c.c.
dQ
d eR L(i
2 Hu )R + c.c. + 1 ( b + M1 B )2
e LH
2
136
CBB
CY Y
CY B
bFB FB +
bFY FY +
bFY FB
M
M
M
D

F
+ b Tr F W F W + b Tr F G F G
M
M
SU(2)
SU(3)
+ d1 BY FY + d2 Y B FB + c1
B C
+ c2
B C
+
+ V (Hu , Hd , b),
(1)
where we have summed over the SU(3) index a = 1, 2, . . . , 8, over the SU(2) index j = 1, 2, 3
G
and over the fermion index i = 1, 2, 3 denoting a given generation. We have denoted with F
W
the field-strength for the gluons and with F the field strength of the weak gauge bosons W .
Y and F B are the field-strengths related to the Abelian hypercharge and the extra Abelian
F
gauge boson, B, which has anomalous interactions with a typical generation of the Standard
Model. The fermions in Eq. (1) are either left-handed or right-handed Dirac spinors fL , fR and
they fall in the usual SU(3)C and SU(2)W representations of the Standard Model. The additional
anomalous U (1)B is accompanied by a shifting Stckelberg axion b. The ci , i = 1, 2, are the
coefficients of the ChernSimons trilinear interactions [19,20] and we have also introduced a
mass term M1 at tree level for the B gauge boson, which is the Stckelberg term. As usual, the
hypercharge is anomaly-free and its embedding in the so-called D-brane basis has been discussed extensively in the previous literature [13,16,24]. Most of the features of the orientifold
construction are preserved, but we do not work with the more general multiple U (1) structure
since our goal is to analyze as close as possible this model making contact with direct phenomenological applications, although our results and methods can be promptly generalized to more
complex situations.
Before moving to the more specific analysis presented in this work, some comments are in
order concerning the possible range of validity of effective actions of this type and the relation
between the value of the cutoff parameter and the Stckelberg mass M1 . This point has been
addressed before in great detail in [21] and we omit any further elaboration, quoting the result.
Lagrangeans containing dimension-5 operators in the form of a WessZumino term may have
a range of validity constrained by M1 g1 g 2 /(64 3 )an , where g1 is the coupling at the chiral vertex where the anomaly an is assigned and g is the coupling constant of the other two
vector-like currents in a typical AVV diagram. More quantitatively, this bound can be reasonably
assumed to be of the order of 105 GeV, by a power-counting analysis. Notice that the arguments
of [21], though based on the picture of partial decoupling of the fermion spectrum, in which
the pseudoscalar field is the phase of a heavier Higgs, remain fully valid in this context (see [21]
for more details). The actual value of M1 is left undetermined, although in the context of string
model building there are suggestions to relate them to specific properties of the compactified
extra dimensions (see for instance [13,16]).
3. The effective action of the MLSOM with a single anomalous U (1)
Having derived the essential components of the classical Lagrangean of the model, now we
try to extend our study to the quantum level, determining the anomalous effective action both
for the Abelian and the non-Abelian sectors, fixing the D, F and C coefficients in front of the
GreenSchwarz terms in Eq. (1). Notice that the only anomalous contributions to San in the
Y -basis before symmetry breaking come from the triangle diagrams depicted in Fig. 1.
Since hypercharge is anomaly-free, the only relevant non-Abelian anomalies to be canceled
are those involving one boson B with two SU(2)W bosons, or two SU(3)C bosons, while the
137
Fig. 1. Anomalous triangle diagrams for the MLSOM.
Abelian anomalies are those containing three U (1) bosons, with the Y 3 triangle excluded by
the hypercharge assignment. These (BSU(2)SU(2)) and (BSU(3)SU(3)) anomalies must be canceled respectively by GreenSchwarz terms of the kind

Db Tr F G F G ,
F b Tr F W F W ,
with F and D to be fixed by the conditions of gauge invariance. In the Abelian sector we have to
focus on the BBB, BY Y and Y BB triangles which generate anomalous contributions that need
to be canceled, respectively, by the GreenSchwarz terms CBB bF B F B , CY Y bF Y F Y and
CY B bF Y F B . Denoting by SY M the anomalous effective action involving the classical nonAbelian terms plus the non-Abelian anomalous diagrams, and with Sab the analogous Abelian
one, the complete anomalous effective action is given by
Seff = S0 + SY M + Sab
with S0 being the classical Lagrangean and

1 ,ij
T
(z, x, y)B (z)Wi (x)Wj (y)
SY M = dx dy dz
2! BW W

1 ,ab
+ TBGG (z, x, y)B (z)Ga (x)Gb (y) ,

2!

Sab =
dx dy dz
(2)
(3)
1
T
(z, x, y)B (z)B (x)B (y)
3! BBB
1
T
(z, x, y)B (z)Y (x)Y (y)
2! BY Y

1
+ TY BB (z, x, y)Y (z)B (x)B (y) .
2!
+
The corresponding 3-point functions, for instance, are given by

,f ,f ,f
,ij
TBW W B Wi Wj = 0|T JB JW i JWj |0B Wi Wj

,f ,f ,f
0|T JB L JW i L JWj L |0B Wi Wj ,
(4)
(5)
and similarly for the others. Here we have defined the chiral currents
,f
JB
,fR
= JB
,fL
+ JB
fR
fL
= gB qB f PR f gB qB f PL f .
(6)
138
Fig. 2. Contributions to a three Abelian gauge boson amplitude before the removal of the Bb gauge boson-Stckelberg
mixing.
The non-Abelian W current being chiral

,f
,fL
JW i JW i
= g2 f i PL f ,
(7)
it forces the other currents in the triangle diagram to be of the same chirality, as shown in Fig. 7.
4. Three gauge boson amplitudes and gauge fixing
4.1. The non-Abelian sector before symmetry breaking
Before we get into the discussion of the gauge invariance of the model, it is convenient to
elaborate on the cancelations of the spurious s-channel poles coming from the gauge-fixing conditions. These are imposed to remove the bB mixing in the effective action. We will perform
our analysis in the basis of the interaction eigenstates since in this basis recovering gauge independence is more straightforward, at least before we enforce symmetry breaking via the Higgs
mechanism. The procedure that we follow is to gauge fix the B gauge boson in the symmetric
phase by removing the Bb mixing (see Fig. 2(C)), so to derive simple Ward identities involving only fermionic triangle diagrams and contact trilinear interactions with gauge bosons. For
this purpose to the Stckelberg term
1
( b + M1 B )2 ,
2
we add the gauge fixing term
1
Lgf = GB2 ,
2
to remove the bilinear mixing, where
1
GB = ( B B M1 b),
B
(8)
(9)
(10)
with a propagator for the massive B gauge boson separated in a gauge independent part P0 and
a gauge dependent one P :

kk
i
i
k k

(11)
g
+
= P0 + P .
2
2
2
2
2
2
k M1
M1
k B M1 M1
We will briefly illustrate here how the cancelation of the gauge dependence due to b and B exchanges in the s-channel goes in this (minimally) gauge-fixed theory. In the exact phase we have
no mixing between all the Y, B, W gauge bosons and the gauge dependence of the B propagator
is canceled by the Stckelberg axion. In the broken phase things get more involved, but essentially the pattern continues to hold. In that case the Stckelberg scalar has to be rotated into its
139
Fig. 3. Unitarity check in SU(2) sector for the MLSOM.
physical component and the two Goldstones GZ and GZ which are linear combinations of G01
and G02 . The cancelation of the spurious s-channel poles takes place, in this case, via the combined exchange of the Z propagator and of the corresponding Goldstone mode GZ . Naturally
the GS interaction will be essential for this to happen.
For the moment we simply work in the exact symmetry phase and in the basis of the interaction
eigenstates. We gauge fix the action to remove the Bb mixing, but for the rest we set the vev
of the scalars to zero. For definiteness let us consider the process W W W W mediated by a
B boson as shown in Fig. 3. We denote by a bold-faced V the BW W vertex, constructed so to
have gauge invariance on the W -lines. This vertex, as we are going to discuss next, requires a
generalized CS counterterm to have such a property on the W -lines. Gauge invariance on the
B-line, instead, which is clearly necessary to remove the gauge dependence in the gauge fixed
action, is obtained at a diagrammatical level by the axion exchange (Fig. 3). The expressions of
the two diagrams are
A + B =

1
i
(L) 2
k VBW W (k1 , k2 ) k VBW W (k1 , k2 ) gB g22 DB
2
2
B M1 M1

4F 2
i

+4
(12)
k1 k2 k1 k2 .
2
2
M
k B M1
k2
Using the equations for the anomalies and the correct value for the GreenSchwarz coefficient F
given in Eq. (62) (and that we will determine in the next section), we obtain
A + B =

2

1
i
(4an k1 k2 ) 4an k1 k2 gB g22 DB(L)
2
2
B M1 M1

an (L) 2
64 M 2
i
+ 2 2 igB g22 DB
k1 k2 k1 k2 ,
2
M M1
k 2 B M12
k2
(13)
so that the cancelation is easily satisfied. The treatment of the SU(3) sector is similar and we
omit it.
4.2. The Abelian sector before symmetry breaking
In the Abelian sector the procedure is similar. For instance, to test the cancelation of the
gauge parameter B in a process BB BB mediated by a B gauge boson we sum the two
gauge dependent contributions coming from the diagrams in Fig. 4 (we consider only the gauge
140
Fig. 4. Unitarity check in Abelian sector for the MLSOM.
dependent part of the s-channel exchange diagrams)

A + B =
2

1
i
4k AAA (k1 , k2 ) 4k AAA (k1 , k2 ) gB3 DBBB
2
2
k 2 B M1 M1

2
i
4
+4
k1 k2 k1 k2 ,
(14)
CBB
2
M
k B M12
and cancelation of the gauge dependences implies that the following identity must hold

2
2
4
16 an 2 3
gB DBBB + 4
CBB = 0,
M
M12 3
(15)
which can be easily shown to be true after substituting the value of the GS coefficient given in
relation (77). In Fig. 5 we have depicted the anomalous triangle diagram BY Y (A) which has to
be canceled by the GreenSchwarz term CMY Y bF Y F Y , that generates diagram (B). In this case
the two diagrams give
A + B =
2

1
i
k VBY Y (k1 , k2 ) k VBY Y (k1 , k2 ) gB gY2 DBY Y
2
2
B M1 M1

2
i
4
+4
k1 k2 k1 k2 .
(16)
CY Y
2
M
k B M12
k2
The condition of unitarity of the amplitude requires the validity of the identity

2
2
16 2
4
2
g
a
g
D
+
4
= 0,
C
n
B
BY
Y
Y
Y
Y
M
M12
(17)
which can be easily checked substituting the value of the GS coefficient CY Y given in relation (78). We will derive the expressions of these coefficients and the factors of all the other
counterterms in the next section. The gauge dependences appearing in the diagrams shown in
Fig. 6 are analyzed in a similar way and we omit repeating the previous steps, but it should be
obvious by now how the perturbative expansion is organized in terms of tree-level vertices and
1-loop counterterms, and how gauge invariance is checked at higher orders when the propagators
of the B gauge boson and of the axion b are both present. Notice that in the exact phase the axion
b is not coupled to the fermions and the pattern of cancelations to ensure gauge independence, in
this specific case, is simplified.
141
At this point we pause to make some comments. The mixed anomalies analyzed above involve
a non-anomalous Abelian gauge boson and the remaining gauge interactions (Abelian/nonAbelian). To be specific, in our model with a single non-anomalous U (1), which is the hypercharge U (1)Y gauge group, these mixed anomalies are those involving triangle diagrams with
the Y and B generators or the B accompanied by the non-Abelian sector. Consider, for instance,
the BY Y triangle, which appears in the Y B Y B amplitude. There are two options that we can
follow. Either we require that the corresponding traces of the generators over each generation
vanish identically

2
1 2 (L)
(e )
qB + (1)2 qB R
Tr qY qB = 2
2

2
2

1 2 (dR )
1
2
(Q )
(u )
= 0,
+ 3 2
qB L +
qB R +
qB
(18)
6
3
3
which can be viewed as a specific condition on the charges of model or, if this is not the case,
we require that suitable one-loop counterterms balance the anomalous gauge variation. We are
allowed, in other words, to fix the two divergent invariant amplitudes of the triangle diagram so
that the corresponding Ward identities for the BY Y vertex and similar anomalous vertices are
satisfied. This is a condition on the parameterization of the Feynman vertex rather than on the
charges and is, in principle, allowed. It is not necessary to have a specific determination of the
charges for this to occur, as far as the counterterms are fixed accordingly. For instance, in the
Abelian sector the diagrams in question are

Y B Y B mediated by Y Tr qY2 qB ,

Y Y Y Y mediated by B Tr qY2 qB ,

BB BB mediated by Y Tr qY qB2 ,
142
YB YB

mediated by B Tr qY qB2 .
(19)
In the MLSOM these traces are, in general, non-vanishing and therefore we need to introduce
defining Ward identities to render the effective action anomaly free.
5. Ward identities, GreenSchwarz and ChernSimons counterterms in the Stckelberg
phase
Having discussed the structure of the theory in the basis of the interaction eigenstates, we
come now to identify the coefficients needed to enforce cancelation of the anomalies in the
1-loop effective action. In the basis of the physical gauge bosons we will be dropping, with this
choice, a gauge dependent (Bb mixing) term that is vanishing for physical polarizations. At the
same time, for exchanges of virtual gauge bosons, the gauge dependence of the corresponding
propagators is canceled by the associated Goldstone exchanges.
Starting from the non-Abelian contributions, the BW W amplitude, we separate the charge/coupling constant dependence of a given diagram from the rest of its parametric structure T using,
in the SU(2) case, the relations
,ij
TBW W B Wi Wj = gB g22

i,j
1
Tr[i j ] Tr qBL T B Wi Wj
8
(L)
1
= gB g22
DB T B Wi Wi ,
2
(20)
fL
(L)
having defined DB = 18 Tr[qBL ] = 18 f qB and T is the 3-point function in configuration
space, with all the couplings and the charges factored out, symmetrized in . Similarly, for the
coupling of B to the gluons we obtain
,ab
2
TBGG B G
a Gb = gB g3

a,b
1
Tr[Ta Tb ] Tr qBL T B G
a Gb
8
(L)
1
= gB g32
DB T B G
a Ga ,
2
a
(21)
while the Abelian triangle diagrams are given by

1
TBBB B B B = gB3 Tr qB3 T B B B = gB3 DBBB T B B B ,

8

1
TBY Y B Y Y = gB gY2 Tr qB qY2 T B Y Y

8
= gB gY2 DBY Y T B Y Y ,

1
TY BB Y B B = gY gB2 Tr qY qB2 T Y B B
8
2
= gY gB DY BB T Y B B ,
(22)
(23)
(24)
with the following definitions for the traces (see also the discussion in Appendices AC)
DBBB =
1 3 1 f R 3 f L 3
qB
,
qB
Tr qB =
8
8
f
(25)
143
Fig. 7. All the anomalous electroweak contributions to a triangle diagram in the non-Abelian sector in the massless
fermion case.
Fig. 8. All the anomalous contributions to a triangle diagram in the Abelian sector for generic vector-axial vector trilinear
interactions in the massless fermion case.
DBY Y =
1 f R f R 2
1
f L f L 2
qB qY
,
qB qY
Tr qB qY2 =
8
8
(26)
1 f R f R 2
1
f L f L 2
qY qB
.
qY q B
Tr qY qB2 =
8
8
(27)
DY BB =
The T vertex is given by the usual combination of vector and axial-vector components
T = TAAA + TAVV + TVAV + TVVA ,

and we denote by (k1 , k2 ) its expression in momentum space

(2)4 (k k1 k2 ) (k1 , k2 ) = dx dy dz eik1 x+ik2 yikz T (z, x, y).
(28)
(29)
We denote similarly with AVV , VAV , VVA the momentum space expressions of the corre
sponding x-space vertices TAVV , TVVA , TVAV , respectively. As illustrated in Figs. 7 and 8, the
complete structure of T is given by

1
(1/2, k1 , k2 ) + (1/2, k2 , k) + (1/2, k, k1 )
3
+ (1/2, k1 , k2 ) + (1/2, k2 , k) + (1/2, k, k1 )

4
= (1/2, k1 , k2 ) + (1/2, k2 , k) + (1/2, k, k1 )
3
= 4AAA ,
(30)
(k1 , k2 ) =
where we have used the relation between the AAA (bold-faced) vertex and the usual vertex,
which is of the form AVV. Notice that
AVV = (1/2, k1 , k2 ),
AVV = (1/2, k2 , k),
AVV = (1/2, k, k1 ),
(31)
144
are the usual vertices with conserved vector current CVC on two lines and the anomaly on a
single axial vertex.
The AAA vertex is constructed by symmetrizing the distribution of the anomaly on each of
the three chiral currents, which is the content of (30). The same vertex can be obtained from the
basic AVV vertex by a suitable shift, with = 1/6, and then repeating the same procedure on
the other indices and external momenta, with a cyclic permutation. We obtain
i 2
(k1 k2 )
4 2 3
and its corresponding anomaly equations are given by
AAA (1/6, k1 , k2 ) = (1/2, k1 , k2 )
an
k1 k2 ,
3
an
k1 AAA (1/6, k1 , k2 ) = k1 k2 ,
3
an
k2 AAA (1/6, k1 , k2 ) =
k2 k1 ,
3
typical of a symmetric distribution of the anomaly.
These identities are obtained from the general shift-relation
(32)
k AAA (1/6, k1 , k2 ) =
(33)
i
(34)
( )
(k1 k2 ) .
4 2
Vertices with conserved axial currents (CAC) can be related to the symmetric AAA vertex in a
similar way
( , k1 , k2 ) = (, k1 , k2 ) +

i 1
AAA (+1/6, k1 , k2 ) = (+1/2, k1 , k2 ) CAC +

(35)
(k1 k2 ) .
4 2 3
At this point we are ready to introduce the complete vertices for this model, which are given
by the amplitude (29) with the addition of the corresponding ChernSimons counterterms where
required. These will be determined later in this section by imposing the conservation of the
SU(2), SU(3) and Y gauge currents. Following this definition for all the anomalous vertices, the
amplitudes can then be written as
1
,aa
2 (L)
SU(3)
VBGG B G
B Ga Ga + c2
B C
,
a Ga = gB g3 DB T
2
1
,ii
(L)
SU(2)
VBW W B Wi Wi = gB g22 DB T B Wi Wi + c1
B C
,
2
VBY Y B Y Y = gB gY2 DBY Y T B Y Y + d1 BY FY ,
VY BB Y B B = gY gB2 DY BB T Y B B + d2 Y B FB ,
VBBB B B B = gB3 DBBB T B B B ,
(36)
which are the anomalous vertices of the effective action, corrected when necessary by suitable
CS interactions in order to conserve all the gauge currents at 1-loop.
Before we proceed with our analysis, which has the goal to determine explicitly the counterterms in each of these vertices, we pause for some practical considerations. It is clear that the
scheme that we have followed in order to determine the structure of the vertices of the effective
action has been to assign the anomaly only to the chiral vertices and to impose conservation of the
145
vector current. There are regularization schemes in the literature that enforce this principle, the
most famous one being dimensional regularization with the t Hooft Veltman prescription for 5
(see also the discussion in part 1). In this scheme the anomaly is equally distributed for vertices
of the form AAA and is assigned only to the axial-vector vertex in triangles of the form AVV
and similar. Diagrams of the form AAV are zero by Furrys theorem, being equivalent to VVV.
We could also have proceeded in a different way, for instance by defining each V, for instance
VBY Y , to have an anomaly only on the B vertex and not on the Y vertices, even if Y has both a
vector and an axial-vector components at tree level and is, indeed, a chiral current. This implies
that at 1-loop the chiral projector has to be moved from the Y to the B vertex by hand, no
matter if it appears on the Y current or on the B current, rendering the Y current effectively
vector-like at 1-loop. This is also what a CS term does. In both cases we are anyhow bond to
define separately the 1-loop vertices as new entities, unrelated to the tree level currents. However,
having explicit ChernSimons counterterms renders the treatment compatible with dimensional
regularization in the t HooftVeltman prescription. It is clear, however, that one way or the other,
the quantum action is not fixed at classical level since the counterterms are related to quantum
effects and the corresponding Ward identities, which force the cancelation of the anomaly to take
place in a completely new way respect to the SM case, are indeed defining conditions on the
theory.
Having clarified this subtle point, we return to the determination of the gauge invariance
conditions for our anomalous vertices. Under B-gauge transformations we have the following
variations (singlet anomalies) of the effective action

(L)
gB g22 4 1
1
B TBW W BW W = i
an B FiW FjW Tr i j DB ,
2!
2! 3 4
2

g
g
1
1
B 3 4
B TBGG BGG = i
an B FaG FbG Tr T a T b DB(L) ,
2!
2! 3 4
and with the normalization given by
1

Tr i j = ij ,
2
we obtain
1

Tr T a T b = ab ,
2
(37)
(38)
(39)
(L)
gB g22 an
1
B TBSU(2)SU(2) BW W = i
B FiW FiW DB ,
(40)
2!
2! 6
(L)
gB g32 an
1
B TBSU(3)SU(3) BGG = i
B FaG FaG DB .
(41)
2!
2! 6
Note, in particular, that the covariantization of the anomalous contributions requires the entire
W and F G
non-Abelian field strengths Fi,
a,
W
W
Fi,
= Wi Wi g2 ij k Wj Wk = Fi,
g2 ij k Wj Wk ,
(42)
G
G
= Ga Ga g3 fabc Gb Gc = Fa,
g3 fabc Gb Gc .
Fa,
(43)
The covariantization of the right-hand side (rhs) of the anomaly equations takes place via higher
order corrections, involving correlators with more external gauge lines. It is well known, though,
that the cancelation of the anomalies in these higher order non-Abelian diagrams (in d = 4) is
only related to the triangle diagram (see [23]).
146
Under the non-Abelian gauge transformations we have the following variations

(L)

gB g22 8 1 B
1
SU(2) TBW W BW W = i
an F Tr F W DB ,
(44)
2!
2! 3 4
(L)

gB g32 8 1 B
1
SU(3) TBGG BGG = i
an F Tr F G DB ,
(45)
2!
2! 3 4
where the hat field strengths F W and F G refer to the Abelian part of the non-Abelian field
strengths W and G. Introducing the notation

1
Tr F W = Tr i j i FjW = i FiW , i, j = 1, 2, 3,
2
a b
G
1
G
Tr F = Tr T T a Fb = a FaG , a, b = 1, 2, . . . , 8,
2
the expressions of the variations become
(46)
(47)
(L)
gB g22 an
1
SU(2) TBW W BW W = i
i F B FiW DB ,
(48)
2!
2! 3
(L)
gB g32 an
1
SU(3) TBGG BGG = i
a F B FaG DB .
(49)
2!
2! 3
We have now to introduce the ChernSimons counterterms for the non-Abelian gauge variations

CS
CS
CS
SU(2)
SU(3)
Snon
(50)
+ c2 B C
,
B C
-Abelian = SBW W + SBGG = c1
with the non-Abelian CS forms given by

1
1
SU(2)
W
C
(51)
= Wi Fi,
+ g2 ij k Wj Wk + cyclic ,
6
3

1
1
SU(3)
G
C
(52)
= Ga Fa,
+ g3 f abc Gb Gc + cyclic ,
6
3
whose variations under non-Abelian gauge transformations are
W

1
(53)
i Fi,
+ cyclic ,
6
G

1
SU(3)
SU(3) C
(54)
+ cyclic .
= a Fa,
6
The variations of the ChernSimons counterterms then become

c1 1 i B
CS
SU(2) SBW
(55)
F FiW ,
W =
2 2

c2 1 a B
CS
F FaG ,
SU(3) SBGG
(56)
=
2 2
and we can choose the coefficients in front of the CS counterterms to obtain anomaly cancelations
for the non-Abelian contributions
2
2
c2 = igB g32 an DB(L) .
c1 = igB g22 an DB(L) ,
(57)
3
3
The variations under B-gauge transformations for the related CS counterterms are then given by
SU(2)
SU(2) C
=
CS
B SBW
W =

c1 1
B FiW FiW ,
2 2
(58)
147

c2 1
(59)
B FaG FaG ,
2 2
where the coefficients ci are given in (57). The variations under the B-gauge transformations for
the SU(2) and SU(3) GreenSchwarz counterterms are respectively given by

F
M1 1
(60)
B b Tr F W F W = F
B FiW FiW ,
M
M 2

M1 1
D
B b Tr F G F G = D
B FaG FaG ,
(61)
M
M 2
and the cancelation of the anomalous contributions coming from the B-gauge transformations
determines F and D as
M
an (L)
an (L)
M
F=
(62)
igB g22 DB ,
D=
igB g32 DB .
M1
2
M1
2
There are some comments to be made concerning the generalized CS terms responsible for the
cancelation of the mixed anomalies. These terms, in momentum space, generate standard trilinear CS interactions, whose momentum structure is exactly the same as that due to the Abelian
ones (see the appendix of part 1 for more details), plus additional quadrilinear (contact) gauge
interactions. These will be neglected in our analysis since we will be focusing in the next sections on the characterization of neutral trilinear interactions. In processes such as Z they
re-distribute the anomaly appropriately in higher point functions.
For the Abelian part Sab of the effective action we first focus on gauge variations on B,
obtaining
CS
B SBGG
=

g3 4 3
1
B TBBB B (z)B (x)B (y) = i B an F B F B B DBBB ,
3!
3! 3 4
2

g
1
1
B gY 4
B TBY Y B (z)Y (x)Y (y) = i
an F Y F Y B DBY Y ,
2!
2! 3 4

gB2 4 2 Y
g
1
Y
B TY BB Y (z)B (x)B (y) = i
an F F B B DY BB ,
2!
2! 3 4
and variations for Y that give

gB gY2 4 2 Y
1
Y TBY Y B (z)Y (x)Y (y) = i
an F F B Y DBY Y ,
2!
2! 3 4
2

g
1
1
Y gB 4
Y TY BB Y (z)B (x)B (y) = i
an F B F B Y DY BB .
2!
2! 3 4
Also in this case we introduce the corresponding Abelian ChernSimons counterterms
CS
CS
CS
= SBY
Sab
Y + SY BB = d1 BY FY + d2 Y B FB
whose variations are given by

d1
CS
Y F B F Y ,
Y SBY
Y =
2

d2
Y SYCSBB = Y F B F B ,
2
and we can fix their coefficients so to obtain the cancelation of the Y -anomaly
2
d1 = igB gY2 an DBY Y ,
3
d2 = igY gB2
an
DY BB .
3
(63)
(64)
(65)
(66)
(67)
(68)
(69)
(70)
(71)
148
Similarly, the gauge variation of B in the corresponding GreenSchwarz terms gives

CBB B
M1
(72)
B bF F B = CBB
B F B F B ,
M
M

M1
CY Y Y
B bF F Y = CY Y
B F Y F Y ,
(73)
M
M

M1
CY B Y
(74)
B bF F B = CY B
B F Y F B
M
M
and on the other hand the B-variations of the fixed CS counterterms are

d1
CS
B F Y F Y ,
B SBY
(75)
Y =
2

d2
B SYCSBB =
(76)
B F Y F B .
2
Finally the cancelation of the anomalous contributions from the Abelian part of the effective
action requires following conditions
M igB3
an DBBB ,
M1 3!
an
M
igB gY2 DBY Y ,
CY Y =
M1
2
a
M
n
igY gB2 DY BB .
CY B =
M1
2
CBB =
(77)
(78)
(79)
Regarding the Y -variations Tr[qB qY2 ] and Tr[qB2 qY ], in general these traces are not identically vanishing and we introduce the CS and GS counterterms to cancel them. Having determined
the factors in front of all the counterterms, we can summarize the structure of the one-loop anomalous effective action plus the counterterms as follows
S = S0 + San + SGS + SCS
1
1
1
= S0 + TBW W BW W + TBGG BGG + TBBB BBB
2!
2!
3!
1
1
+ TBY Y BY Y + TY BB Y BB
2!
2!
CY Y
CY B
CBB
bFB FB +
bFY FY +
bFY FB
+
M
M
M
D

F
+
b Tr F W F W +
b Tr F G F G
M
M
+ d1 BY FY + d2 Y B FB

SU(2)
SU(3)
+ c1
B C
+ c2
B C
,
(80)
where S0 is the classical action. At this point we are ready to define the expressions in momentum
space of the vertices introduced in Eq. (36), denoted by V, obtaining
1 i 2
1
VBGG = 4 DB(L) gB g32 AAA (+1/6, k1 , k2 ) + DB(L) gB g32

(k1 k2 ) ,
2
2 2 3
1 i 2
1 (L)
(L)
(k1 k2 ) ,
VBW W = 4 DB gB g22 AAA (+1/6, k1 , k2 ) + DB gB g22

2
2 2 3
(81)
(82)
i 2
(k1 k2 ) ,
2 3
i 1
= 4DY BB gY gB2 AAA (+1/6, k1 , k2 ) DY BB gY gB2 2

(k1 k2 ) ,
3
VBY Y = 4DBY Y gB gY2 AAA (+1/6, k1 , k2 ) + DBY Y gB gY2
VY BB
VBBB = 4DBBB gB3 AAA (+1/6, k1 , k2 ),
149
(83)
(84)
(85)
where for the generalized CS terms we consider only the trilinear CS interactions whose momentum structure is the same as the Abelian ones as already discussed in Section 5. The factor
1/2 overall in the non-Abelian vertices comes from the trace over the generators. These vertices
satisfy standard Ward identities on the external Standard Model lines, with an anomalous Ward
identity only on the B-line
k1 VBY Y (k1 , k2 ) = 0,
(86)
k2 VBY Y (k1 , k2 ) = 0,
k VBY Y (k1 , k2 ) = 4DBY Y gB gY2 an

k1 k2 ,
(87)
(88)
and obviously the B-currents contain the total anomaly an = 2i 2 . The same anomaly equations
given above for VBY Y holds for the VBGG and VBW W vertices but with a 1/2 factor overall. The
anomaly equations for the Y BB vertex are
an
k1 k2 ,
2
an
k2 VY BB (k1 , k2 ) = 4DY BB gY gB2

k2 k1 ,
2
k1 VY BB (k1 , k2 ) = 4DY BB gY gB2
k VY BB (k1 , k2 ) = 0,
(89)
(90)
(91)
where the chiral current Y has to be conserved so to render the 1-loop effective action gauge
invariant. Introducing a symmetric distribution of the anomaly, in the BBB case the analogous
equations are
an
k1 k2 ,
3
an
k2 VBBB (k1 , k2 ) = 4DBBB gB3

k2 k1 ,
3
3 an
k1 k2 .
k VBBB (k1 , k2 ) = 4DBBB gB
k1 VBBB (k1 , k2 ) = 4DBBB gB3
(92)
(93)
(94)
A study of the issue of the gauge dependence in these types of models can be found in [23].
Clearly, this study is more involved, but the cancelations of the gauge dependent terms in specific
classes of diagrams can be performed both in the exact phase and in the broken phase, having
re-expressed the fields in the basis of the mass eigenstates. The approach that we follow is then
clear: we worry about the cancelation of the anomalies in the exact phase, having performed a
minimal gauge fixing to remove the B mixing with the axion b, then we rotate the fields and
re-parameterize the Lagrangean around the non-trivial vacuum of the potential. We will see in
the next sections that with this simple procedure we can easily discuss simple basic processes
involving neutral and charged currents exploiting the invariance of the effective action under
re-parameterizations of the fields.
150
6. The neutral currents sector in the MLSOM

In this section we move toward the phenomenological analysis of a typical process which
exhibits the new trilinear gauge interactions at 1-loop level. As we have mentioned in the introduction, our goal here is to characterize this analysis at a more formal level, leaving to future
work a numerical study. It should be clear, however, from the discussion presented in this and in
the next sections, how to proceed in a more general case. The theory is well-defined and consistent so that we can foresee accurate studies of its predictions for applications at the LHC in the
future.
We proceed with our illustration starting from the definition of the neutral current in the model,
which is given by

LN C = f g2 W3 T 3 + gY Y AY + gB YB B f ,
(95)
that we express in the two basis, the basis of the interaction eigenstates and of the mass eigenstates. Clearly in the interaction basis the bosonic operator in the covariant derivative becomes
F g2 W3 T 3 + gY Y AY + gB YB B = gZ QZ Z + gZ QZ Z + eQA ,
(96)
where Q = T 3 + Y . The rotation in the photon basis gives

A
A
A

W3 = OW
A + OW
Z + OW
Z ,
3
3Z
3Z
= OYA A
A
B = OBZ
Z
AY
+ OYAZ Z + OYAZ Z ,
A

+ OBZ
Z
(97)
(98)
(99)
and performing the rotation on F we obtain

A
3
A
A
3
A
A
F = A g2 OW
g
T
+
g
O
Y
+
Z
O
T
+
g
O
Y
+
g
O
Y
Y
2
Y
B
B
Z
Y
W
Y
Z
BZ
3
3

A
3
A
A
T
+
g
O
Y
+
g
O
+ Z g2 OW

Y
B
YZ
BZ YB ,
3Z
(100)
where the electromagnetic current can be written in the usual way

A
A g2 OW
T 3 + gY OYA Y = eA Q,
3
(101)
with the definition of the electric charge as

A
= gY OYA =
e = g2 OW
3
gY g2
(102)
gY2 + g22
Similarly for the neutral Z current we obtain

A
A
Z g2 OW
T 3 + gY OYAZ Y + gB OBZ
YB
3Z

A
A
gY OYAZ + gY OYAZ Q + gB OBZ
YB
= Z T 3 g2 OW
3Z

A
gY OYAZ
gB OBZ
3
= Z gZ T +
Q+
YB ,
A
A
g2 OW
gY OYAZ
g2 OW
gY OYAZ
3Z
3Z
(103)
where we have defined

A
gZ = g2 OW
gY OYAZ g =
3Z
g2
.
cos W
(104)
151
We can easily work out the structure of the covariant derivative interaction applied on a lefthanded or on a right-handed fermion. For this reason it is convenient to introduce some notation.
We define
gY OYAZ
sin2 W ,
gZ
A
gB OBZ
gB
Z

1 so that
B =
gZ
2
Z
Q=
(105)
lim Z
B = 0,
(106)
M1
and similarly for the Z neutral current

A
A
gZ = g2 OW
gY OY Z ,
3Z
Z
Q =
gY OYAZ
,
gZ
Z
B =
A
gB OBZ

.
gZ
(107)
We can easily identify the generators in the (Z, Z , A ) basis. These are given by
3L
L
Z L
Z R
Z R
R
L
Z =Q
+ Z
Q
Z + QZ = T
Q Q + B YB + Q Q + B YB ,

L
Z L
Z R
Z R
Z = Q R + Q
L = T 3L + Z
Q
Q Q + B YB + Q Q + B YB ,
Z
Z
R,
=Q
L +Q
Q
(108)
Q
Z, Q
Z ). To express a given correlator, say ZA A in the
which will be denoted as Qp = (Q,
Q
Z, Q
Z ) the generators in
(W3 , AY , B) basis we proceed as follows. We denote with Qp = (Q,

the photon basis (A , Z, Z ) and with gp = (e, gZ , gZ ) the corresponding couplings. Similarly,
Qp = (T 3 , Y, YB ) are the generators in the interaction basis (W3 , AY , B) and gp = (g2 , gY , gB )
the corresponding couplings, so that

Z Z + eQA
gZ Q Z Z + gZ Q
LN C =

g2 T 3 W3 + gY Y AY + gB YB B .
=
(109)
7. The Z vertex in the Standard Model
Before coming to the computation of this vertex in the MLSOM we first start reviewing its
structure in the SM.
We show in Fig. 9 the Z vertex in the SM, where we have separated the QED contributions
from the remaining corrections RW . This vertex vanishes at all orders when all the three lines
are on-shell, due to the LandauYang theorem. A direct proof of this property for the fermionic
1-loop corrections has been included in Appendices AC, where we show the on-shell vanishing
of the vertex.
The QED contribution contains the fermionic triangle diagrams (direct plus exchanged) and
the contributions in RW include all the remaining ones at 1-loop level. In this case the separation between the pure QED contributions (due to the 2 fermionic diagrams) and the remaining
corrections, which are separately gauge invariant on the photon lines, is rather straightforward,
though this is not the case, in general, for more complicated electroweak amplitudes. Specifically,
as shown in Fig. 10, RW , contains ghosts, Goldstones and all other exchanges. An exhaustive computation of all these contributions is not needed for the scope of this discussion and
will be left for future work. We have omitted diagrams of the type shown in Figs. 11, 12.
These are removed by working in the R gauge for the Z boson. Notice, however, that even
without a gauge fixing these decouple from the anomaly diagrams in the massless fermion
152
Fig. 9. The Z vertex to lowest order in the Standard Model, with the anomalous contributions and the remaining weak
corrections shown separately.
Fig. 10. Some typical electroweak corrections, involving the charged Goldstones (here denoted by G, ghosts contributions
(u ) and W exchanges.
Fig. 11. Z G0Z mixing in the broken phase in the SM.
limit since the Goldstone does not couple to massless fermions. In Fig. 13 we show how the
anomaly is re-distributed in an AAA diagram by a CS interaction, generating an AVV vertex.
To appreciate the role played by the anomaly in this vertex we perform a direct computation
of the two anomaly diagrams and include the fermionic mass terms. A direct computation gives
e2 g f 2
gA Qf
cos W
f

1
1
1
d 4p
5
Tr
p
/ mf
p
/ k/ mf p
/ k/1 mf
(2)4
+ (k1 k2 , ),
(110)
G (k, k1 , k2 ) =
which can be cast in the form
153
Fig. 12. Same as in Fig. 11 but for the MLSOM.
Fig. 13. Re-distribution of the anomaly via the CS counterterm.
1x
1
1
f
e2 g
2
(k, k1 , k2 ) = 2
gA Qf dx1
dx2
2 cos W
f

1
(1 x1 x2 )(x2 k1 x1 k2 ) k2 k1 + k1 k2

+ (1 x1 x2 )
k1 k2 x2 k1 x1 k2 + ( )

+
k1 k2 x2 (x2 x1 1)k1 x1 (x2 x1 + 1)k2 ,
(111)
where
= m2f + x2 (x2 1)k12 + x1 (x1 1)k22 2x1 x2 k1 k2 ,
f
(112)
and we have introduced the gZ,A and gZ,V couplings of the Z with
1 f
1 f
f
f
gZ,A = T3 ,
(113)
gZ,V = T3 Qf sin2 W .
2
2
This form of the amplitude is obtained if we use the standard Rosenberg definition of the anomalous diagrams and it agrees with [29]. In this case the Ward identities on the photon lines are
defining conditions for the vertex. Naturally, with the standard fermion multiplet assignment the
anomaly vanishes since
f
(114)
gA Q2f = 0.
f
Because of the anomaly cancelation, the fermionic vertex is zero also off-shell, if the masses of
all the fermions in each generation are degenerate, in particular if they are massless. Notice that
this is not a consequence of the LandauYang theorem.
Let us now move to the Ward identity on the Z-line. A direct computation gives
154
k G = (k1 + k2 ) G

1x
1
1
f
1
e2 g
1
2
2
= 2
gA Qf
k1 k2
dx2
mf dx1
.
2

cos W
f
(115)
The presence of a mass-dependent term on the right-hand side of (115) constitutes a break-down
of axial current conservation for massive fermions, as expected.
7.1. The Z vertex in anomalous Abelian models: The HiggsStckelberg phase
The presence of anomalous generators in a given vertex renders some trilinear interactions
non-vanishing also for massless fermions. In fact, as we have shown in the previous section, in the
SM the anomalous triangle diagrams vanish if we neglect the masses of all the fermions, and this
occurs both on-shell and off-shell. The only left over corrections are related to the fermion mass
and these will also vanish (off-shell) if all the fermions of a given generation are mass degenerate.
The on-shell vanishing of the same vertices is a consequence of the structure of the amplitude,
as we show in Appendices AC. The extraction of the contribution of the anomalous generators
in the trilinear vertices can be obtained starting from the 1-particle irreducible effective action,
written in the basis of the interaction eigenstates, and performing the rotation of the trilinear
interaction that project onto the Z vertex.
In order to appreciate the differences between the SM result and the analogous one in the
anomalous extensions that we are considering, we start by observing that only in the Stckelberg
phase (M1 = 0 and vu = vd = 0) the anomaly-free traces vanish,

Y Y Y gY3 Tr Q3Y = 0,

Y W3 W3 gY g22 Tr QY T 3 T 3 = 0,
(116)
because of charge assignment. A similar result is valid also in the HS phase if the Yukawa couplings are neglected. Coming to extract the Z vertex we rotate the anomalous diagrams of the
effective action into the mass eigenstates, being careful to separate the massless from the massive fermion contributions. Hence, we split the Y Y Y vertex into its chiral contributions and
performing the rotation of the fields we get the following contributions

1
Y Y Y gY3 Tr Q3Y
3!
1
3
3
1
=
+ gY3 QR
gY3 QL
Y,f LLL
Y,f RRR
8
8
f
R 2
1
1
+ gY3 QL
+ gY3 QL
QR QL LRL
Y,f QY,f LRR
8
8 Y,f Y,f Y,f
2 R
L 2
1
1
+ gY3 QL
+ gY3 QR
Y,f QY,f LLR
Y,f QY,f RLL
8
8

R 2 L
31 R
L
R
31
+ gY QY,f QY,f QY,f RLR

+ gY QY,f QY,f RRL
8
8
1
RY Y Y +
Z A
A
3!
(117)
155
where the dots indicate all the other projections of the type ZZ , Z , etc. Here LLL,
RLR, etc., indicate the (clockwise) insertion of L/R chiral projectors on the vertices
of the anomaly diagrams.
For the Y W W vertex the structure is more simple because the generator associated to W3 is
left-chiral
2
1
Y W W gY g22 Tr QY T 3
2!

3 2
3 2
21 L
21 R
gY g2 QY,f TL,f LLL

+ gY g2 QY,f TL,f RLL
=
8
8
f
1 YWW
R
+ .
2!
The BY Y vertex works in same way of Y Y Y ,
Z A
A
(118)

1
BY Y gB gY2 Tr QB Q2Y
2!

L 2
R 2
1
1
gB gY2 QL
+ gB gY2 QR
=
B,f QY,f LLL
B,f QY,f RRR
8
8
f
R 2
1
1
+ gB gY2 QL
+ gB gY2 QL
QR QL LRL
B,f QY,f LRR
8
8 B,f Y,f Y,f
L 2
1
1
Q
+ gB gY2 QL
QL QR LLR + gB gY2 QR
RLL
8 B,f Y,f Y,f
8 Y,f Y,f

1
L
R
21 R
R
L
Q
+ gB gY2 QR
Q
Q
RLR
+
g
g
Q
Q
RRL
B
Y
8 B,f Y,f Y,f
8 B,f Y,f Y,f
1 BY Y
R
Z A
+ .
A
2!
Finally, BW W vertex is similar to Y W W
(119)
2
1
BW W gY g22 Tr QB T 3
2!

3 2
3 2
1
21 R
gB g22 QL
Q
T
T
LLL
+
g
g
RLL
=
B
2
8 B,f L,f
8 B,f L,f
f
Z A
A
1 BW W
R
+
2!

2

R Y Y Y = 3 O AT 22 O AT 21 ,

2

R Y W W = 2 O AT 11 O AT 12 O AT 21 + O AT 11 O AT 22 ,
2

R BY Y = O AT 21 O AT 32 ,
2

R BW W = O AT 11 O AT 32 .
(120)
(121)
which are the product of rotation matrices that project the anomalous effective action from the
interaction eigenstate basis over the Z, gauge bosons.
156
We have expressed the generators in their chiral basis, and their mixing is due to mass insertions over each fermion line in the loop. The ellypsis refers to additional contributions which do
not project on the vertex that we are interested in but which are present in the analysis of the
remaining neutral vertices, ZZ , Z , etc. The notation O AT indicates the transposed of the
rotation matrix from the interaction to the mass eigenstates. To obtain the final expression of the
amplitude in the interaction eigenstate basis one can easily observe that in the helicity conserving amplitudes LLL and RRR the mass dependence in the fermion loops is all contained
in the denominators of the propagators, not in the Dirac traces. The only diagrams that contain
a mass dependence at the numerators are those involving chirality flips (LLR, RRL) which
contribute with terms proportional to m2f . These terms contribute only to the invariant amplitudes
A1 and A2 of the Rosenberg representation [23] and, although finite, they disappear once we impose a Ward identity on the two photon lines, as requested by CVC for the two photons. A similar
result is valid for the SM, as one can easily figure out from Eq. (111). Therefore, the amplitudes
can be expressed just in terms of LLL and RRR correlators, and since the mass dependence is
at the denominators of the propagators, one can easily show the relation
LLL = RRR
(122)
valid for any fermion mass mf . Defining LLL LLL (mf = 0), we can express the only
independent chiral graph as sum of two contributions
LLL (mf = 0) = LLL (0) + LLL (mf )
(123)
where we define
LLL (0) LLL (mf = 0),
LLL (mf ) LLL (mf = 0) LLL (mf = 0).
(124)
Also, one can verify quite easily that
LLL (0) = AV V (0) + V AV (0) + V V A (0) + AAA (0) = 4AAA (0).
(125)
A second contribution to the effective action comes from the 1-loop counterterms containing generalized CS terms. There are two ways to express these counterterms: either as separate
3-linear interactions or as modifications of the two invariant amplitudes of the Rosenberg parameterization A1 , A2 . These amplitude depend linearly on the momenta of the vertex [23]. For
instance we use
AAA (0)
an
(k1 k2 ) = AV V (0),
(126)
which allows to absorb completely the CS term, giving conserved Y/W3 currents in the interaction eigenstate basis. In this case we move from a symmetric distribution of the anomaly in the
AAA diagram, to an AV V diagram. These currents interpolate with the vector-like vertices (V)
of the AVV graph.
Notice that once the anomaly is moved from any vertex involving a Y/W3 current to a vertex
with a B current, it is then canceled by the GS interaction. The extension of this analysis to the
complete mf -dependent case for LLL (mf = 0) is quite straightforward. In fact, after some rearrangements of the Z amplitude, we are left with the following contributions in the physical
157
basis in the broken phase

1
AV V (mf = 0) gY3 fY Y Y R Y Y Y + gY g22 fY W W R Y W W
Z |mf =0 =
4
f
+ gB gY2 fBY Y R BY Y + gB g22 fBW W R BW W Z A

A
(127)
where we have defined the anomalous chiral asymmetries as

L 2
R 2
3 2

,
fBW W = QL
QR
fBY Y = QL
B,f QY,f
B,f QY,f
B,f TL,f .
(128)
The conditions of gauge invariance force the coefficients in front of the CS terms to be
1 BY Y
1 BW W
f ,
DBW W =
f
,
DBY Y =
8
8
f
(129)
which have been absorbed and do not appear explicitly, while the SM chiral asymmetries are
defined as
3 R 3
3 2

fY Y Y = QL
(130)
fY W W = QL
QY,f ,
Y,f
Y,f TL,f ,
and the triangle AV V (mf = 0) is given as in (111). Notice that Eq. (127) is in complete agreement with the SM result shown in (111), obtained by removing the contributions proportional
to the B gauge bosons and setting the chiral asymmetries of Y and W3 to zero. In particular, if
the gauge bosons are not anomalous and in the chiral limit (mf = 0 or mf = m) this trilinear
amplitude vanishes.
As we have already pointed out, the amplitude for the Z process is espressed in terms of
6 invariant amplitudes that can be easily computed and take the form
AV V = A1 (k1 , k2 )
[k1 , , , ] + A2 (k1 , k2 )
[k2 , , , ] + A3 (k1 , k2 )
[k1 , k2 , , ]k1
+ A4 (k1 , k2 )
[k1 , k2 , , ]k2 + A5 (k1 , k2 )
[k1 , k2 , , ]k1
+ A6 (k1 , k2 )
[k1 , k2 , , ]k2 ,
(131)
with
A1 (k1 , k2 ) = k1 k2 A3 (k1 , k2 ) + k22 A4 (k1 , k2 ),
A2 (k1 , k2 ) = A1 (k2 , k1 ),
A5 (k1 , k2 ) = A4 (k2 , k1 ),
A6 (k1 , k2 ) = A3 (k2 , k1 ).
(132)
Also A1 (k1 , k2 ) = A1 (k2 , k1 ) as one can easily check by a direct computation. We obtain
1
A3 (k1 , k2 ) =
2
1x

dx
dy
A4 (k1 , k2 ) =
1
2
1
0
1x

dx
dy
0
y(1 y)k12
xy
,
+ x(1 x)k22 + 2xyk1 k2 m2f
x(1 x)
.
y(1 y)k12 + x(1 x)k22 + 2xyk1 k2 m2f
(133)
The computation of these integrals can be done analytically and the various regions 0 < s < 4m2f ,
mf s/2, and mf 0 can be studied in detail. In the case of both photons on-shell, for
158
Fig. 14. Two photon processes initiated by a q q annihilation with a Z exchange.
instance, and s > 4m2f we obtain

m2f
m2f
1
2
2

Li2
Li2
,
2s
s
s
1 1 4m2f /s
1 + 1 4m2f /s

1 4m2f /s
1
1
ArcTanh
A4 (k1 , k2 ) = +
.
s
s
1 4m2 /s
A3 (k1 , k2 ) =
(134)
Notice that the case in which the two photons are on-shell and light fermions are running in the
loop, then the evaluation of the integral requires particular care because of infrared effects which
render the parameteric integrals ill-defined. The situation is similar to the case of the coupling
of the axial anomaly to on-shell gluons in spin physics [30], when the correct isolation of the
massless quarks contributions is carried out by moving off-shell on the external lines and then
performing the mf 0 limit.
7.2. q q with an intermediate Z
In this section we are going to describe the role played by the new anomaly cancelation mechanism in simple processes which can eventually be studied with accuracy at a hadron collider
such as the LHC. A numerical analysis of processes involving neutral currents can be performed
along the lines of [9] and we hope to return to this point in the near future. Here we intend to
discuss briefly some of the phenomenological implications which might be of interest. Since the
anomaly is canceled by a combination of ChernSimons and GreenSchwarz contributions, the
study of a specific process, such as Z , which differs from the SM prediction, requires, in
general, a combined analysis both of the gauge sector and of the scalar sector.
We start from the case of a quarkantiquark annihilation mediated by a Z that later undergoes
a decay into two photons. At leading order this process is at parton level described by the annihilations of a valence quark q and a sea antiquark q from the two incoming hadrons, both of them
collinear and massless. In Fig. 14 we have depicted all the diagrams by which the process can
take place to lowest order. Radiative corrections from the initial state are accurately known up to
next-to-next-to-leading order, and are universal, being the same of the DrellYan cross section.
In this respect, precise QCD predictions for the rates are available, for instance around the Z
resonance [9].
159
Fig. 15. Gluon fusion contribution to double-photon production. Shown are also the scalar exchanges (B) and (D) that
restore gauge invariance and the axi-Higgs exchange (E).
Fig. 16. The q q annihilation channel (A, B). Scalar exchanges in the neutral sector involving the two Higsses and the
axi-Higgs (C, D, E).
In the SM, gauge invariance of the process requires both a Z gauge boson exchange and
the exchange of the corresponding Goldstone GZ , which involves diagrams (A) and (B). In
the MLSOM a direct GreenSchwarz coupling to the photon (which is gauge dependent) is
accompanied by a gauge independent axion exchange. If the incoming quarkantiquark pair is
massless, then the Goldstone has no coupling to the incoming fermion pair, and therefore (B)
is absent, while gauge invariance is trivially satisfied because of the massless condition on the
fermion pair of the initial state. In this case only diagram (A) is relevant. Diagram (B) may also
be set to vanish, for instance in suitable gauges, such as the unitary gauge. Notice also that the
triangle diagrams have a dependence on mf , the mass of the fermion in the loop, and show two
contributions: a first contribution which is proportional to the anomaly (mass independent) and a
correction term which depends on mf .
As we have shown above, the first contribution, which involves an off-shell vertex, is absent
in the SM, while it is non-vanishing in the MLSOM. In both cases, on the other hand, we have
mf dependent contributions. It is then clear that in the SM the largest contribution to the process
comes from the top quark circulating in the triangle diagram, the amplitude being essentially
proportional only to the heavy top mass. On the Z resonance and for on-shell photons, the cross
section vanishes in both cases, as we have explained, in agreement with the LandauYang theorem. We have checked these properties explicitly, but they hold independently of the perturbative
order at which they are analyzed, being based on the Bose symmetry of the two photons. The
cross section, therefore, has a dip at Q = MZ , where it vanishes, and where Q2 is the virtuality
of the intermediate s-channel exchange.
160
An alternative scenario is to search for neutral exchanges initiated by gluongluon fusion. In

this case we replace the annihilation pair with a triangle loop (the process is similar to Higgs
production via gluon fusion), as shown in Fig. 15. As in the decay mechanism discussed above,
the production mechanism in the SM and in the MLSOM are again different. In fact, in the
MLSOM there is a massless contribution appearing already at the massless fermion level, which
is absent in the SM. The production mechanism by gluon fusion has some special features as
well. In ggZ production and Z decay, the relevant diagrams are (A) and (B) since we need
the exchange of a GZ to obtain gauge invariance. As we probe smaller values of the Bjorken
variable x, the gluon density raises, and the process becomes sizable. On the other hand, in a
pp collider, although the quark annihilation channel is suppressed since the antiquark density is
smaller than in a p p collision, this channel still remains rather significant. We have also shown
in this figure one of the scalar channels, due to the exchange of a axi-Higgs.
Other channels such as those shown in Fig. 16 can also be studied, these involve a lepton pair
in the final state, and their radiative corrections also show the appearance of a triangle vertex.
This is the classical DrellYan process, that we will briefly describe below. In this case, both the
total cross section and the rapidity distributions of the lepton pair and/or an analysis of the charge
asymmetry in s-channel exchanges of W s would be of major interest in order to disentangle the
anomaly inflow. At the moment, errors on the parton distributions and scale dependences induce
indeterminations which, just for the QCD background, are around 4% [9], as shown in a high
precision study. It is expected, however, that the statistical accuracy on the Z resonance at the
LHC is going to be a factor 100 better. In fact this is a case in which the experiment can do better
than the theory.
7.3. Isolation of the massless limit: The Z amplitude
The isolation of the massless from the massive contributions can be analized in the case of
resolved photons in the final state. As we have already mentioned in the prompt photon case
the amplitude, on the Z resonance, vanishes because of Bose symmetry and angular momentum
conservation. We can, however, be on the Z resonance and produce one or two off-shell photons
that undergo fragmentation. Needless to say, these contributions are small. However, the separation of the massless from the massive case is well defined. One can increase the rates by asking
just for 1 single resolved photon and 1 prompt photon. Rates for this process in pp-collisions
have been determined in [31]. We start from the case of off-shell external photons of virtuality s1
and s2 and an off-shell Z (Z ). Following [32], we introduce the total vertex V (k1 , k2 , mf ),
which contains both the massive mf dependence (corresponding to the triangle amplitude .
Its massless counterpart V (0) V (k1 , k2 , mf = 0), obtained by sending the fermion mass to
zero. The Rosenberg vertex and the V vertex are trivially related by a Schoutens transformation,
moving the index from the Levi-Civita tensor to the momenta of the photons
V (k1 , k2 , mf ) = A(k1 , k2 , mf )[, , , k2 ]s1 A(k2 , k1 , mf )[, , , k1 ]s2
+ A(k1 , k2 , mf )[, , k1 , k2 ]k1 + A(k2 , k1 , mf )[, , k2 , k1 ]k2

B(k1 , k2 , mf )[, , k1 , k2 ]k ,
with k k1 k2 = 0 and si = ki2 (i = 1, 2), and
(135)

1
1
1
A(k1 , k2 , mf ) =
(s s1 + s2 )
(s + s2 ) + (6/)ss1 s2 #1
2
2

1
+ s2
(3/)s(s s1 s2 ) #2
2

+ ss2 + m2f + (3/)ss1 s2 (s s1 + s2 ) C#0 ,
B(k1 , k2 , mf ) =

1 1
1
(s s1 s2 ) + s1
+ (3/)s2 (s + s1 s2 ) #1
2
2

1
+ s2
+ (3/)s1 (s s1 + s2 ) #2
2

+ s1 s2 m2f + (3/)ss1 s2 (s s1 s2 ) C#0 ,
161
(136)
(137)
with
= (s, s1 , s2 ),
(138)
being the usual Mandelstam function and where the analytic expressions for #i and C#0 are
given by
ai + 1
a3 + 1
a3 ln
(i = 1, 2),
ai 1
a3 1

3
1
bi 1
bi 1
C#0 =
Li2
Li2
a i + bi
a i bi
i=1

bi + 1
bi + 1
Li2
,
+ Li2
a i bi
a i + bi
#i = ai ln
(139)
and
ti = si i
,
= (t1 , t2 , t3 ),
ai =
1 + (2mf )2 /ti (i = 1, 2, 3),
b1 = (t1 t2 t3 )/ or cyclic.
(140)
For mf = 0 the two expressions above become

#i = ln(ti /t3 ) (i = 1, 2),

1
C#0 = (1/ ) 2 (2) Li2 (x1 ) Li2 (x2 ) + Li2
+ ln x1 ln x2 ,
x3
(141)
with
xi =
(bi + 1)
(bi 1)
(i = 1, 2, 3).
(142)
These can be inserted into (136) and (137) together with mf = 0 to generate the corresponding
V (0) vertex needed for the computation of the massless contributions to the amplitude.
162
With these notations we clearly have

= V (k1 , k2 , mf ),
(0) = V (k1 , k2 ),
(mf ) = V (k1 , k2 , mf ) V (k1 , k2 ).
(143)
7.4. Extension to Z
To isolate the contribution to the decay on the resonance, we keep one of the two photons
off-shell (resolved). We choose s1 = 0, and s2 virtual. We denote by the corresponding
vertex in this special kinematical configuration. The Z boson is on-shell. In this case at one-loop
the result simplifies considerably [33]

= F2 s2
[, , , k1 ] + k2
[, , k1 , k2 ] ,
(144)
with F2 expressed as a Feynman parametric integral
1
F2 =
2 2
1
dz1 dz2 dz3 (1 z1 z2 z3 )
0
m2f
z2 z3
.
z2 z3 s2 z1 z3 MZ2
(145)
Setting F2 F (z, rf ) where f (z, r) is a dimensionless function of

rf = m2Z /4m2f ,
z = s2 /MZ2 ,
(146)
and for vanishing mf (rf = MZ2 /4m2f ), the corresponding massless contribution is expressed as F (z, ) with, in general
F (z, rf ) =

1
I (rf z, rf ) I (rf , rf ) + 1 z ,
2
4(1 z)
where
I (x, rf ) = 2

2
x 1
1
ln x + 1 x
ln x + 1 x
x
rf
(147)
for x < 0
1 x 1
1 1 2
sin
x+
x
for 0 < x < 1
sin
x
rf

i
i 2

x 1
1
=2
,
ln x + x 1
ln x + x 1
x
2
rf
2
for x > 1.
(148)
=2
The mf = 0 contribution is obtained in the rf + limit,

1
(ln z + 1 z) for z > 0,
4(1 z)2

1
=
ln |z| + i + 1 z for z < 0.
2
4(1 z)
F (z, ) =
(149)
In these notations, the infinite fermion mass limit (mf or r 0), gives F (z, 0) = 0 and
we find
163
= = F (z, rf ),
(0) = (0) = F (z, ),
(mf ) = (mf ) = F (z, rf ) F (z, ),
(150)
which can be used for a numerical evaluation. The decay rate for the process is given by

1
(Z ) =
d 4 k1 d 4 k2 k12 k22 Q2 |MZ |2 2()4 (k k1 k2 ),
4MZ
(151)
where

|MZ |2 = AZ Z AZ Q
,
Z = g +
kk
,
MZ2
+
Q
= g
kk
.
Q2
(152)
We have indicated with Q the virtuality of the photon. A complete evaluation of this expression,
to be of practical interest, would need the fragmentation functions of the photon (see [31] for
an example). A detailed analysis of these rates will be presented elsewhere. However, we will
briefly summarize the main points involved in the analysis of this and similar processes at the
LHC, where the decay rate is folded with the (NLO/NNLO) contribution from the initial state
using QCD factorization.
Probably one of the best way to search for neutral current interactions in hadronic collisions at
the LHC is in lepton pair production via the DrellYan mechanism. QCD corrections are known
for this process up to O(s2 ) (next-to-next-to-leading order, NNLO), which can be folded with the
NNLO evolution of the parton distributions to provide accurate determinations of the hadronic pp
cross sections at the 4% level of accuracy [9]. The same computation for DrellYan can be used
to analize the pp Z process since the WV (hadronic) part of the process is universal,
with WV defined below. An appropriate (and very useful) way to analyze this process would be
to perform this study by defining the invariant mass distribution

d
= Z Q2 , MV2 WV , Q2 ,
2
dQ
(153)
where = Q2 /S, which is separated into a pointlike contribution Z

V Q2 , MV2 =
(Z )
.
2 cos 2 N (Q2 M 2 )2 + M 2 2
4MZ sin W
W c
Z
Z Z
(154)
and a hadronic structure functions WZ . This is defined via the integral over parton distributions
and coefficient functions ij

WZ Q2 , MZ2
1
=
1
dx1
1
dx2

dx ( xx1 x2 )P DijV x1 , x2 , 2f ij x, Q2 , 2f ,
(155)
where f is the factorization scale. The choice f = Q, with Q the invariant mass of the
pair, removes the log(Q/M) for the computation of the coefficient functions, which is, anyhow,
164
arbitrary. The non-singlet coefficient functions are given by

(0)
q q = (1 x),
(1)
q q

S (MV2 )
log(1 x)
CF (1 x) 8 (2) 16 + 16
=
4
1x
+

2
1+x
log x ,
8(1 + x) log(1 x) 4
1x
(156)
with CF = (Nc2 1)/(2Nc ) and the + distribution is defined by

log(1 x)
1x
log(1 x)
(1 x)
= (1 x)
1x
+
1
log(1 x)
,
dx
1x
(157)
while at NLO appears also a qg sector

S (MV2 )
(1 x)2
2
2
2
1
+
2x
T
+
1
7x
(1)
=
2x
log
+
6x
.
F
qg
4
x
(158)
Other sectors do not appear at this order. Explicitly one gets

WZ Q2 , MZ2 =
i
1
dx1
1
dx ( xx1 x2 )
dx2
0

qi x1 , 2f qi x2 , 2f + qi x1 , 2f qi x2 , 2f q q x, Q2 , 2f

+ qi x1 , 2f g x2 , 2f + qi x2 , 2f g x1 , 2f qg x, Q2 , 2f ,
(159)
where the sum is over the quark flavours. The identification of the generalized mechanism of
anomaly cancelation requires that this description be extended to NNLO, which is now a realistic
possibility. It involves a slight modification of the NNLO hard scatterings known at this time and
an explicit computation is in progress.
8. Conclusions
We have presented a study of a model inspired by the structure encountered in a typical string
theory derivation of the Standard Model. In particular we have focused our investigation on the
characterization of the effective action and worked out its expression in the context of an extension containing one additional anomalous U (1). Our analysis specializes and, at the same
time, extends a previous study of models belonging to this class. The results that we have presented are generic for models where the Stckelberg and the Higgs mechanism are combined
and where an effective Abelian anomalous interaction is present. Our analysis has then turned
toward the study of simple processes mediated by neutral current exchanges, and we have focused, specifically, on one of them, the one involving the Z vertex. In particular our findings
clearly show that new massless contributions are present at 1-loop level when anomalous generators are involved in the fermionic triangle diagrams and the interplay between massless and
massive fermion effects is modified respect to the SM case. The typical processes considered in
our analysis deserve a special attention, given the forthcoming experiments at the LHC, since
165
they may provide a way to determine whether anomaly effects are present in some specific reactions. Other similar processes, involving the entire neutral sector should be considered, though
the two-photon signal is probably the most interesting one phenomenologically.
Given the high statistical precision (0.05% and below on the Z peak, for 10 f b1 of integrated
luminosity) which can be easily obtained at the LHC, there are realistic chances to prove or
disprove theories of these types. Concerning the possibility of discovering extra anomalous Z ,
although there are stringent upper bounds on their mixing(s) with the Z gauge boson, it is of
outmost importance to bring this type of analysis even closer to the experimental test by studying
in more detail the peculiarities of anomalous gauge interactions for both the neutral and the
charged sectors along the lines developed in this work. This analysis is in progress and we hope
to report on it in the near future.
Acknowledgements
We dedicate this work to the memory of Hidenaga Yamagishi, remembering his remarkable scientific talent, his outstanding human qualities and his unique and inspirational love for
physics.
We thank Elias Kiritsis for having brought the topics discussed in this work to our attention
and Theodore Tomaras, Marco Roncadelli, Marco Guzzi, Roberta Armillis, Andrea Spirito and
Antonio Quintavalle for discussions. The work of C.C. was supported in part by the European
Union through the Marie Curie Research and Training Network Universenet (MRTN-CT-2006035863). He thanks the Theory Group at the University of Liverpool and in particular Alon
Faraggi for discussions and for the kind hospitality. N.I. was partially supported by the European
contract MRTN-CT-2004-512194.
Appendix A. A summary on the single anomalous U (1) model
We summarize in this appendix some results concerning the model with a single anomalous
U (1) discussed in the main sections. These results specialize and simplify the general discussion
of [19] to which we refer for further details. We will use the hypercharge values
f
qY
QL
1/6
uR
2/3
dR
1/3
L
1/2
eR
1
dR
(d )
qB R
L
(L)
qB
eR
(e )
qB R
R
0
and general U (1)B charge assignments

f
qB
QL
(Q )
qB L
uR
(u )
qB R
The covariant derivatives act on the fermions fL , fR as

(f )
D fL = + iA + iql L gl Al, fL ,

(f )
D fR = + iA + iql R gl Al, fR ,
R
( )
qB R
(A.1)
with l = Y, B Abelian index, where A is a non-Abelian Lie algebra element and write the lepton
doublet as

Li
.
Li =
(A.2)
eLi
166
We will also use standard notations for the SU(2)W and SU(3)C gauge bosons
i
W = Wi = i Wi , with i = 1, 2, 3,
2
a a
G = G = Ta Ga , with a = 1, 2, . . . , 8,
2
with the normalizations
1
1

Tr T a T b = ab .
Tr i j = ij ,
2
2
The interaction Lagrangean for the leptons becomes

a
Li
lep
Lint = Li g2 Wa gY qY(L) AY gB qB(L) B
eLi
eLi
2

(eR ) Y
(eR )
+ eRi gY qY A gB qB B eRi

( )
( )
+ Ri gY qY R AY gB qB R B Ri .
(A.3)
(A.4)
(A.5)
(A.6)
As usual we define the left-handed and right-handed currents

1
1
J J5 ,
JR = J + J5 ,
2
2
J = JR + JL ,
J5 = JR JL .
JL =
Writing the quark doublet as

uLi
,
QLi =
dLi
(A.7)
(A.8)
we obtain the interaction Lagrangean

a a
i i
uLi
quarks
(QL ) Y
(QL )
Z
Lint = ( uLi d Li ) g3 G g2 W gY qY A gB qB B
dLi
2
2

(u )
(u )
+ u Ri gY qY R AY gB qB R B uRi

(d )
(d )
+ dRi gY qY R AY gB qB R B dRi .
(A.9)
As we have already mentioned in the introduction, we work with a 2-Higgs doublet model,
and therefore we parameterize the Higgs fields in terms of 8 real degrees of freedom as
+
+
Hd
Hu
Hu =
(A.10)
,
Hd =
,
0
Hu
Hd0
where Hu+ , Hd+ and Hu0 , Hd0 are complex fields. Specifically
+
HuR
H + iH
+ iH +
Hd = dR dI ,
uI ,
2
2
Hu = Hu+ ,
Hd+ = Hd .
Hu+ =
(A.11)
Expanding around the vacuum we get for the uncharged components

Hu0 = vu +
0 + iH 0
HuR
uI ,
2
Hd0 = vd +
0 + iH 0
HdR
dI .
2
(A.12)
167
The Weinberg angle is defined via cos W = g2 /g, sin W = gY /g, with
g 2 = gY2 + g22 .
(A.13)
We also define cos = vd /v, sin = vu /v and

v 2 = vd2 + vu2 .
(A.14)
The mass matrix in the mixing of the neutral gauge bosons is given by
W3

Lmass = W3 AY B M2 AY ,
B
(A.15)
where
g2 2 v 2
1
2
M =
g2 gY v 2
4
g2 xB
g2 gY v 2
gY 2 v 2
gY x B
g2 xB
,
gY x B
2M12 + NBB
(A.16)
with

NBB = quB2 vu2 + qdB2 vd2 gB2 ,

xB = quB vu2 + qdB vd2 gB .
(A.17)
(A.18)
The orthonormalized mass squared eigenstates corresponding to this matrix are given by
A
O11
OA =
12
A
O13
gY
g22 +gY2
g2
g22 +gY2
(A.19)

g2 (2M12 g 2 v 2 +NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )

g 2 x 4+ g2 (2M 2 g 2 v 2 +N +(2M 2 g 2 v 2 +N )2 +4g 2 x 2 )2
B
BB
BB
B
1
1
2
g 4 xB
A

O21
2
2
2
2
2
2
2
2
2
gY (2M1 g v +NBB + (2M1 g v +NBB ) +4g xB )
OA =

22
g2
2
2
2
2 2
2 2
2
2 2 2
A
g xB 4+ g4 xB2 (2M1 g v +NBB + (2M1 g v +NBB ) +4g xB )
O23

2

2
4+ 4g 2 (2M12 g 2 v 2 +NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )2
(A.20)
g xB
One can see that these results reproduce the analogous relations of the SM in the limit of very
large M1
g2
,
g
g xB
g
A
O23

1
2
2 M1
2
A
lim O21
=
M1
A
lim O22
=
M1
so that
gY
,
g
A
lim O23
= 0.
M1
168
Similarly, for the other matrix elements of the rotation matrix O A we obtain

g2 (2M12 +g 2 v 2 NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )

2
2
g xB 4+ 4g 2 (2M12 +g 2 v 2 NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )2
g xB
A

O31
gY (2M12 +g 2 v 2 NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )

OA =

32
2 g2
g xB 4+ 4 2 (2M12 +g 2 v 2 NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )2
A
O33
g xB
2

4+
g2
2
g 4 xB
A
lim O31
= 0,
(A.21)
(2M12 +g 2 v 2 NBB + (2M12 g 2 v 2 +NBB )2 +4g 2 xB2 )2
whose asymptotic behavior is described by the limits

g2 xB
g2
g Y xB
gY
A
A

1 ,
O32

1 ,
O31
2 M12
2
2 M12
2
M1
A
lim O32
= 0,
A
O33
1,
A
lim O33
= 1.
M1
M1
(A.22)
(A.23)
These mass-squared eigenstates correspond to one zero mass eigenvalue for the photon A , and
two non-zero mass eigenvalues for the Z and for the Z vector bosons, corresponding to the mass
values

2
1
m2Z = 2M12 + g 2 v 2 + NBB 2M12 g 2 v 2 + NBB + 4g 2 xB2
4

g2v2
1 g 2 xB2
1 g 2 xB2

(A.24)
2
+ 4
NBB g 2 v 2 ,
2
M1 4
M1 8

2
1
m2Z = 2M12 + g 2 v 2 + NBB + 2M12 g 2 v 2 + NBB + 4g 2 xB2
4
NBB
.
M12 +
(A.25)
2
The mass of the Z gauge boson gets corrected by terms of the order v 2 /M1 , converging to the
SM value as M1 , with M1 the Stckelberg mass of the B gauge boson, the mass of the Z
gauge boson can grow large with M1 .
The physical gauge fields can be obtained from the rotation matrix O A
A
W3
Z = O A AY ,
(A.26)
Z
B
which can be approximated at the first order as
g2
gY
g
O A gg2 + O(
12 )
vs g22
1
ggY + O(
12 )
gY
2
1
0
g
2
1
(A.27)
1 + O(
12 )
The mass squared matrix (A.16) can be diagonalized as
0 0
A

T

A 2
A
0 m2
(A ZZ )O M O
Z = (A ZZ )
Z
Z
0 0
0
A
0 Z .
Z
m2Z
(A.28)
169
It is straightforward to verify that the rotation matrix O A satisfies the proper orthogonality relation
T
O A O A = 1.
(A.29)
A.1. Rotation matrix O on the axi-Higgs
This matrix is needed in order to rotate into the mass eigenstates of the CP odd sector, relating
the axion and the two neutral Goldstones of this sector to the Stckelberg field b and the CP
odd phases of the two Higgs doublets
Im Hu0
Im H 0 = O G01 .
(A.30)
d
b
G02
We refer to [19] for a detailed discussion of the scalar sector of the model, where, in the presence
of explicit phases (PQ breaking terms), the mass of the axion becomes massive from the massless
case. The PQ symmetric contribution is given by

2
VPQ =
2a Ha Ha + aa Ha Ha
a=u,d

2

2ud Hu Hu Hd Hd + 2ud HuT 2 Hd ,
while the PQ breaking terms are

i(quB qdB ) Mb
i(quB qdB ) Mb 2
1 + 1 H Hd e
1
VP/Q/ = b1 Hu Hd e
u
b
B
B

i(qu qd ) M
1
+ 2 Hu Hu Hu Hd e
b
B
B

i(qu qd ) M
1 + c.c.,
+ 3 Hd Hd Hu Hd e
where b1 has mass squared dimension, while 1 , 2 , 3 are dimensionless. Defining

b1 2
+ 2 tan ,
c = 4 41 + 3 cot + 2
v sin 2
(A.31)
(A.32)
(A.33)
and using vd = v cos , vu = v sin together with

cot =
cos
vd
= ,
sin
vu
tan =
sin
vu
= ,
cos
vd
(A.34)
from the scalar potential [19] one can extract the mass eigenvalues of the model for the scalar
sector. The mass matrix has 2 zero eigenvalues and one non-zero eigenvalue that corresponds to
a physical axion field, , with mass

B

qu qdB v sin 2 2
(quB qdB )2 vu2 vd2
1
1
2
2
2
= c v 1 +
. (A.35)
m = c v 1 +
2
M1
2
2
v2
M12
The mass of this state is positive if c < 0. Notice that the mass of the axi-Higgs is the result of
two effects: the presence of the Higgs vevs and the presence of a PQ-breaking potential whose
parameters can be small enough to drive the mass of this particle to be very light. We refer to
[23] for a simple illustration of this effect in an Abelian model. In the case of a single anomalous
U (1) O can be simplified as shown below.
170
Introducing N given by
1
N=
(quB qdB )2
M12
1+
=
vd2 vu2
v2
1+
(quB qdB )2
M12
(A.36)
v 2 sin2 2
4
and defining
(quB qdB )
(q B qdB )
vu = u
v sin ,
M1
M1
1
N1 =
,
1 + Q21
Q1 =
O the following matrix
N cos
O =
N sin
vsNQ1 cos
sin
cos
0
(A.37)
(A.38)
1 cos
N 1 Q
1 sin ,
N 1 Q
N 1
(A.39)
where we defined
Q 1 = Q1 cos
(A.40)
and
1
1
=
N 1 =
1 + Q 21
1 + Q21 cos2
1
=
1+
(quB qdB )2
M12
=
1+
v 2 sin2 cos2
(quB qdB )2
M12
vu2 vd2
v2
(A.41)
One can see from (A.36) that N 1 = N , and the explicit elements of the 3-by-3 rotation matrix
O can be written as

1
1
N = N cos ,
O 11 =
(A.42)
=

vu vuvvd
(quB qdB )
M12
v2
vu
B 2 2 2 +1
B
M1
(qu qd ) vu vd

21
(quB qdB )
M1

O 31 =

vd
M12
v2
(quB qdB )2 vu2 vd2
+1
1
N = N sin ,
vd vuvvd
(A.43)
1
M12
v2
(quB qdB )2 vu2 vd2
+1
1

=
M1
v
(quB qdB )vu u
(quB qdB )2
M12

vu
= sin ,
O 12 =
vu2 + vd2
= N Q1 cos ,
+
(A.44)
v2
vu2 vd2
(A.45)

22

32

13
=
vd
vu2
+ vd2
= cos ,
= 0,
1+

(A.46)
=
(quB qdB )2
M12
vu2 vd2
v2
(quB

qdB ) vu vd2
M1
v2

(quB qdB )
1 cos ,
=N
vu cos cos = N Q
M1

(q B qdB ) vu2 vd
1
O 23 =
u
M1
v2
(q B q B )2 v 2 v 2
1 + u 2d uv 2 d

= N

33
171
(A.47)
(A.48)
M1

qdB )
1 sin ,
vu cos sin = N Q
M1
1
= N.
(quB
=
1+
(quB qdB )2 vu2 vd2

v2
M12
It can be easily checked that this is an orthogonal matrix

T
O = 133 .
O
(A.49)
(A.50)
(A.51)
A.2. Vanishing of the amplitude for on-shell external physical states

An important property of the triangle amplitude is its vanishing for on-shell external physical
states.
The vanishing of the amplitude for on-shell physical states can be verified once we have
assumed conservation of the vector currents. This is a simple example of a result that, in general,
goes under the name of the LandauYang theorem. In our case we use only the expression of
the triangle in the Rosenberg parameterization [34] and its gauge invariance to obtain this result.
We stress this point here since if we modify the Ward identity on the correlator, as we are going
to discuss next, additional interactions are needed in the analysis of processes mediated by this
diagram in order to obtain consistency with the theorem.
We introduce the 3 polarization four-vectors for the , , and lines, denoted by e, 1 and
2 respectively, and we use the Sudakov parameterization of each of them, using the massless
vectors k1 and k2 as a longitudinal basis on the light-cone, plus transversal () components
which are orthogonal to the longitudinal ones. We have
e = (k1 k2 ) + e ,
1 = ak1 + 1 ,
2 = bk2 + 2 ,
(A.52)
where we have used the condition of transversality e k = 0, 1 k1 = 0, 2 k2 = 0, the external

lines being now physical. Clearly e k1 = e k2 = 0, and similar relations hold also for 1
and 2 , all the transverse polarization vectors being orthogonal to the light-cone spanned by k1
and k2 . From gauge invariance on the lines in the invariant amplitude, we are allowed to drop
the light-cone components of the polarizators for these two lines
e 1 2 = e 1 2 ,
(A.53)
172
and a simple computation then gives (introducing e (0, e) and similar)
e 1 2 = a 1
[k1 k2 , 1 , 2 , e]

= a 1
k1 k2 , 1 , 2 , (k1 k2 ) + e
( 1 2 ) e = 0,
(A.54)
since the three transverse polarizations are linearly dependent. Notice that this proof shows that
Z with all three particles on-shell does not occur. As usual one needs extreme care when
massless fermions are running in the loop. The situation is analogous to that encountered in spin
physics in the analysis of the EMC result, where the puzzle was resolved [30] by moving to the
massless fermion case starting from off-mass shell external lines.
Appendix B. Massive versus massless contributions
Here we briefly discuss the computation of the mass contributions to the amplitude. We start
from the massless fermion limit. The anomaly coefficient in (20) can be obtained starting from
the triangle diagram in momentum space. For instance we get
,ij
BSU(2)SU(2)
f L L

= gB g22 Tr i j
qB
f
fL 3

= gB g22 Tr i j
qB (i)
q k/) PL (/
q k/1 ) PLq/ ]
d 4 q Tr[ PL (/
4
2
2
(2)
q (q k1 ) (q k)2
+ (k1 k2 , )
1 fL 3

qB (i)
= gB g22 Tr i j
8
f

q k/) (1 5 )(/
q k/1 ) (1 5 )/
q]
d 4 q Tr[ (1 5 )(/
4
2
2
2
(2)
q (q k1 ) (q k)
+ (k1 k2 , )
(B.1)
and isolating the four anomalous contributions of the form AAA, AVV, VAV and VVA we obtain
DBL =
1 fL
1 f L
qB .
Tr qB
8
8
(B.2)
Similarly we obtain
BBB
= gB3

f R 3
qB
R + gB3
= gB3

f R 3
qB

(i)3

f L 3
qB
L
Tr[ PR (/
q k/) PR (/
q k/1 ) PR q/ ]
(2)4
q 2 (q k1 )2 (q k)2
d 4q
+ gB3

f L 3
qB

(i)3
173
q k/) PL (/
q k/1 ) PLq/ ]
d 4 q Tr[ PL (/
4
2
2
(2)
q (q k1 ) (q k)2
+ (k1 k2 , )
1 f R 3 3
qB
(i)
= gB3
8

q k/) (1 + 5 )(/
q k/1 ) (1 + 5 )/
q]
d 4 q Tr[ (1 + 5 )(/
4
2
2
2
(2)
q (q k1 ) (q k)
f L 3
1
qB (i)3
+ gB3
8
Tr[ (1 5 )(/
q k/) (1 5 )(/
q k/1 ) (1 5 )/
q]
4
2
2
2
(2)
q (q k1 ) (q k)
+ (k1 k2 , ),
1 fL
1 f L
qB ,
DBL = Tr qB
8
8
d 4q
1
1 3
Tr qB =
8
8
DBBB =
f R 3
qB
f L 3
.
qB
(B.3)
The other coefficients reported in Eq. (27) are obtained similarly.

Appendix C. CS and GS terms rotated
The rotation of the CS and GS terms into the physical fields and the goldstone gives
2

BY Y
VCS
= d1 BY FY = (i)d1 (k1 k2 ) O AT 21 O AT 32 Z A
A + ,

BW W
Abelian
VCS
= c1 B C
2

= (i)c1 (k1 k2 ) O AT 11 O AT 32 Z A
A + ,
bY Y
VGS
=
CY Y
CY Y
k k Y Y
bFY FY = 4
b
M
M

2
CY Y
=4
k k O31 O AT 21 A
A
M

2

+ O32 C1 + O33 C1 O AT 21 GZ A

A + ,
bW W
VGS
=
F
CY Y b
k k Wi Wi
bT r[FW FW ] = 4
M
M 2

2
F
= 4 k k O31 O AT 11 A
A
M

2

+ O32 C1 + O33 C1 O AT 11 GZ A

A + .
(C.1)
These vertices appear in the cancelation of the gauge dependence in s-channel exchanges of Z
gauge bosons in the R gauge. The dots refer to the additional contributions, proportional to
interactions of , the axi-Higgs, with the neutral gauge bosons of the model.
174
References
[1] M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 648 (2003) 215;
J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 187.
[2] C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 147.
[3] P. Binetruy, P. Ramond, Phys. Lett. B 350 (1995) 49.
[4] P. Binetruy, E. Dudas, Phys. Lett. B 389 (1996) 503;
G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 (1996) 3728.
[5] P. Binetruy, G. Dvali, Phys. Lett. B 388 (1996) 241.
[6] H. Georgi, J.E. Kim, H.P. Nilles, Phys. Lett. B 437 (1998) 325.
[7] M. Quiros, G. von Gersdorff, Phys. Rev. D 68 (2003) 105002.
[8] C.T. Hill, Phys. Rev. D 71 (2005) 046002;
C.T. Hill, C.K. Zachos, Phys. Rev. D 73 (2006) 085001.
[9] A. Cafarella, C. Corian, M. Guzzi, JHEP 0708 (2007) 030.
[10] R. Blumenhagen, D. Lust, S. Steiberger, hep-th/0610327, and references therein.
[11] M. Bertolini, M. Billo, A. Lerda, J. Morales, R. Rousso, Nucl. Phys. B 743 (2006) 1;
P. Anastasopoulos, T.P.T. Dijkstra, E. Kiritsis, A.N. Schellekens, Nucl. Phys. B 759 (2006) 83.
[12] A.F. Faraggi, S. Forste, C. Timirgaziu, JHEP 0608 (2006) 057.
[13] L.E. Ibanez, R. Rabadan, A. Uranga, Nucl. Phys. B 542 (1999) 112.
[14] I. Antoniadis, E. Kiritsis, J. Rizos, Nucl. Phys. B 637 (2002) 92.
[15] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, vols. I, II, Cambridge Univ. Press, 1987.
[16] D.M. Ghilencea, L.E. Ibanez, N. Irges, F. Quevedo, JHEP 0208 (2002) 016;
D.M. Ghilencea, Nucl. Phys. B 648 (2003) 215.
[17] P. Anastasopoulos, JHEP 0308 (2003) 005;
P. Anastasopoulos, Phys. Lett. B 588 (2004) 119.
[18] E. Kiritsis, P. Anastasopoulos, JHEP 0205 (2002) 054.
[19] C. Corian, N. Irges, E. Kiritsis, Nucl. Phys. B 746 (2006) 77.
[20] P. Anastasopoulos, M. Bianchi, E. Dudas, E. Kiritsis, JHEP 0611 (2006) 057;
P. Anastasopoulos, hep-th/0701114.
[21] J. Preskill, Ann. Phys. 210 (1991) 323.
[22] C. Corian, N. Irges, hep-ph/0612140.
[23] C. Corian, N. Irges, S. Morelli, JHEP 0707 (2007) 008.
[24] I. Antoniadis, E. Kiritsis, J. Rizos, T. Tomaras, Nucl. Phys. B 660 (2003) 81.
[25] B. Kors, P. Nath, Phys. Lett. B 586 (2004) 366.
[26] B. Kors, P. Nath, Phys. Lett. B 586 (2004) 366;
B. Kors, P. Nath, JHEP 0412 (2004) 005;
B. Kors, P. Nath, JHEP 0507 (2005) 069;
D. Feldman, Z. Liu, P. Nath, Phys. Rev. Lett. 97 (2006) 021801;
D. Feldman, Z. Liu, P. Nath, JHEP 0611 (2006) 007;
D. Feldman, Z. Liu, P. Nath, hep-ph/0702123.
[27] T.J. Marshall, D.G.C. McKeon, hep-th/0610034.
[28] A. Mirizzi, G.R. Raffelt, P. Serpico, arXiv: 0704.3044 [astro-ph];
E. Mass, X. Redondo, AIP Conf. Proc. 878 (2006) 387394;
A. De Angelis, O. Mansutti, M. Roncadelli, arXiv: 0707.4312 [astro-ph];
A. De Angelis, O. Mansutti, M. Roncadelli, arXiv: 0707.2695 [astro-ph];
A. Dupays, M. Roncadelli, Nucl. Phys. B (Proc. Suppl.) 168 (2007) 44;
S. Abel, J. Jaeckel, V.V. Khoze, A. Ringwald, hep-ph/0608248;
H. Gies, J. Jaeckel, A. Ringwald, Phys. Rev. Lett. 97 (2006) 140402.
[29] A. Barroso, F. Boudjema, J. Cole, N. Dombey, Z. Phys. C 28 (1983) 149.
[30] R.D. Carlitz, J.C. Collins, A.H. Mueller, Phys. Lett. B 214 (1988) 229.
[31] C. Corian, L.E. Gordon, Nucl. Phys. B 469 (1996) 202.
[32] B.A. Kniehl, J.H. Kuhn, Nucl. Phys. B 329 (1990) 547.
[33] K. Hagiwara, T. Kuruma, Y. Yamada, Nucl. Phys. B 369 (1992) 171.
[34] L. Rosenberg, Phys. Rev. 129 (1963) 2786.
The AdS5 S 5 superstring quantum spectrum

from the algebraic curve
Nikolay Gromov a,b, , Pedro Vieira a,c
a Laboratoire de Physique Thorique de lEcole Normale Suprieure et lUniversit Paris-VI, Paris 75231, France
b St. Petersburg INP, Gatchina, 188 300 St. Petersburg, Russia
c Departamento de Fsica e Centro de Fsica do Porto, Faculdade de Cincias da Universidade do Porto,
Rua do Campo Alegre, 687, 4169-007 Porto, Portugal

Abstract
We propose a method for computing the energy level spacing around classical string solutions in AdS5
S 5 . This method is based on the integrable structure of the string and can be applied to an arbitrary classical
configuration. Our approach treats in equal footing the bosonic and fermionic excitations and provides an
unambiguous prescription for the labeling of the fluctuation frequencies. Finally we revisit the computation
of these frequencies for the SU(2) and SL(2) circular strings and compare our results to the existing ones.
1. Introduction
In this paper we will study the semiclassical quantization of the AdS5 S 5 MetsaevTseytlin
superstring [1]. As a warm-up let us consider a one-dimensional non-relativistic particle in a
smooth potential. In terms of the quasi-momenta
p(x)
h (x)
,
i (x)
* Corresponding author at: Laboratoire de Physique Thorique de lEcole Normale Suprieure et lUniversit Paris-VI,
Paris 75231, France.
E-mail addresses: nikgromov@gmail.com (N. Gromov), pedrogvieira@gmail.com (P. Vieira).
doi:10.1016/j.nuclphysb.2007.07.032
176
N. Gromov, P. Vieira / Nuclear Physics B 789 (2008) 175208
Fig. 1. Analytical structure of a quasi-momenta p(x) of a one-dimensional system. Left: for low lying states p(x) is a
collection of poles. Right: for high energy states the poles condense into a square root branch cut.
the Schrdinger equation for the wave function takes the Riccati form

p 2 i hp
= 2m(E V ).
(1)
What do we know about p(x)? It is an analytical function which has, by definition, a pole with
residue
h
=
(2)
i
at each of the zeros of the wave function. For the N th excited state we will have N poles. On the
other hand, for very excited states, the right-hand side in (1) is much larger than h and

p pcl 2m(E V )
describes now a two-sheet Riemann surface. What happened was that, as N , the poles in
p(x) started to be denser and denser, condensing in a square root cut (see Fig. 1). Thus, in the
semiclassical limit we retrieve the BohrSommerfeld quantization

1
1
(3)
pcl (z) dz
p(z) dz = N,
2 h
2 h
C
where C encircles the cut. The first integral is precisely the action variable of the classical motion.
To anticipate the forthcoming notation we name such integrals filling fractions.
When we consider more degrees of freedom, in particular when we move to higher dimensions, let us say two, the situation is not just a little worse. Indeed, we have no proper generic
recipe, except from lattice calculations, to extract the quantum spectrum, or a part of it, of an
interacting quantum field theory. However, if we are lucky, it might happen that the theory is
integrable. If it is the case, we can identify the action variables, apply the BohrSommerfield
condition and find the quasi-classical spectrum of the theory.
For a wide class of two-dimensional sigma models this happens to be the case and the procedure is known explicitly. The central object is a collection of quasi-momenta, pi (x), whose
derivative defines a many-sheet Riemann surface. These sheets can be connected by several cuts,
to each of which we can associate a filling fraction by integrating the quasi-momenta around the
cut as in (3). These are the action variables of the theory. Grosso modo, these filling fractions
measure the size of the cut. Finally, when going through these cuts the quasi-momenta can jump
by 2n with n being an integer labeling the cut.
The superstring on AdS5 S 5 background falls into this class of theoriesthe model is known
to be classically integrable [2,3], the algebraic curve was built [4], and thus one can expect to
quasi-classically quantize the string. In the string language, when we choose which Riemann
sheets we connect by a cut we are choosing which string polarization, i.e., which degree of
freedom, to excite. The number n and the filling fraction associated to the cut are in strict analogy
with the mode number and amplitude of a Fourier mode in a free theory such as the string in flat
space [4].
177
It is now widely believed that the integral equations describing the quasi-momenta pi (x) are
the continuous limit of a system of discrete equations carrying the name of Bethe ansatz equations
[5,6]. As in the example we considered, the quasi-momenta are given by a sum of poles which
condense into cuts in the classical limit [7].
Going back to our simple example, we can see that the existence of such discrete equations is
2 2
indeed highly natural. For that purpose let us consider a simple harmonic oscillator, V = m2 x .
From (1) it follows that p(x) = imx + O(1/x). Since the quasi-momentum is a meromorphic
function with N poles on the real axis, it must be given by
h 1
.
i
x xi
N
p(x) = imx +
i=1
Then, from the large x behavior in (1) we read immediately

1
E = h N +
,
2
while from the cancelation of each of the xi poles in the same equation we get1
h 1
,
2m
x i xj
N
xi =
(4)
j =i
which strongly resembles the equations one finds in the Bethe ansatz context.
At this point let us leave this instructive example behind and resume our main discussion.
When we expand the superstring action around some classical solution, characterized by some
conserved charges, we obtain, for the oscillations, a quadratic Lagrangian whose quantization
yields, for the semiclassical spectrum,

E = Ecl +
(5)
NA,n EA,n ,
A,n
where we have dropped the zero energy excitation and denoted the number of quanta with energy
EA,n by NA,n . The subscript A labels the several possible string polarizations we can excite
while the mode number n is the Fourier mode of the quantum fluctuation. In this article we shall
address the question of finding this quasi-classical spectrum for the AdS5 S 5 superstring using
the algebraic curve mentioned above.
Let us explain the idea behind the computation. There are basically two main steps involved.
First we construct the curve associated with the classical trajectory around which we want to
consider the quantum fluctuations. This will be given by some Riemann surface with some cuts
uniting some of the eight sheets. The second step consist of considering the small excitations
around this classical solution in the spirit of [8]. In terms of the algebraic curve this consists of
adding some microscopic cuts to this macroscopic background. By microscopic cuts we mean
some finite number of poles whose residue we know [5,6,9], just like in the simple example (2).
Then, by construction, the energy of the perturbed configuration is quantized as in (5). We must
stress that the knowledge of the residue, of utmost importance, is indeed the only extra input we
needed to compute the quasi-classical spectrum.

1 Its solution is given by the zeros of the Hermite polynomials, H (

N
2m x ) = 0.
i
h
178
As an application of this method we compute the fluctuation frequencies around the circular
su(2) and sl(2) string. These solutions belong to a family of circular solutions whose quasimomenta we computed explicitly in Appendix A. The frequencies we compute in this way were
obtained in [10,11] and [12,13] by direct analysis of the fluctuations in the MetsaevTseytlin GS
superstring action. In the end of Section 5 we will propose a precise prescription for the labeling
of the fluctuation frequencies and discuss briefly the computation of the 1-loop shift.
2. Coset model and algebraic curve
In this section we shall review the construction of the algebraic curve of Beisert et al. [4]
describing the classical motion of the superstring in the target superspace
PSU(2, 2|4)
,
SP(2, 2) SP(4)
whose bosonic part is AdS5 S 5 . The MetsaevTseytlin superstring action is of GS type, formulated as a supercoset [1].
2.1. The supercoset
This section follows closely [14]. The matrix superalgebra su(2, 2|4) is spanned by the 8 8
supertraceless supermatrices
A B
M=
,
C D
where A and B belong to u(2, 2) and u(4), respectively, while the fermionic components are
related by

I22
0
C = B
.
0
I22
The psu(2, 2|4) superalgebra is the quotient of this algebra by the matrices proportional to the
identity. Since the su(2, 2|4) algebra enjoys the automorphism
0 1 0 0

T
T
EA E EC E
1 0 0 0
, E=
M =
,
EB T E ED T E
0 0 0 1
0 0 1 0
such that 4 = 1, the algebra
is endowed with a Z4 grading. This means that any algebra element
can be decomposed into 3i=0 M (i) , where M (n) = i n M (n) . Explicitly
M (0,2) =
M (1,3) =
0
1 AEAT E
2
0
DED T E

.
0
BiEC T E
1
2 CiEB T E
0
(6)
We see that the M (0) elements span, by definition, the denominator algebra sp(2, 2) sp(4) of
the coset. Then, the remaining bosonic elements, M (2) , orthogonal to the former, generate the
(orthogonal) complement of sp(2, 2) sp(4) in su(2, 2) su(4).
179
Finally, the MetsaevTseytlin action for the GS superstring in AdS5 S 5 is then given in
terms of the algebra current
J = g 1 dg,
(7)
where g(, ) is a group element of PSU(2, 2|4), by

str J (2) J (2) J (1) J (3) + str J (2) ,

S=
4
(8)
where the last term ensures that J (2) is supertraceless. Besides the obvious global PSU(2, 2|4)
left multiplication symmetry the action (8) possesses a local gauge symmetry, g gH with
H SP(2, 2) SP(4), under which
J (i) H 1 J (i) H,
i = 1, 2, 3,
while J (0) transforms as a connection. The equations of motion following from (8) are equivalent
to the conservation of the Noether current associated with the global left multiplication symmetry
d k = 0,
(9)
where k = gKg 1 and K = J (2) + 12 J (1) 12 J (3)

For a purely bosonic representative g we can write
g=
1
2
Q 0
,
0 R
where R SU(4) and Q SU(2, 2). Then we see that RERT is a good parametrization of
SU(4)/SP(4) = S 5 ,
because, by definition, it is invariant under R RH with H SP(4). In the same way QEQT
describes the AdS space. Then we can define the embedding coordinates u and v by the simple
relations
uj jS = RERT ,
v j jA = QEQT ,
(10)
where S , A are the gamma matrices of SO(6) and SO(4, 2). By construction these coordinates
will automatically satisfy
1 = u26 + u25 + u24 + u23 + u22 + u21 ,
1 = v62 + v52 v42 v32 v22 v12 .
Then the bosonic part of the action can be expressed in the usual non-linear model form
2

Sb =
d d h h u u + u (u u 1) (u v) .
4
0
(11)
180
Fig. 2. A possible analytical stricture of the quasi-momenta of an integrable sigma model. Many types of cuts are now
possible. Cuts can join different sheets and each cut is marked by its mode number nij . In flat space limit they become
numbers of fourier modes. The number of microscopical poles constituting the given cut is called a filling fraction and
can be calculated as a contour integral (3).
2.2. The algebraic curve

As follows from the equations of motion and the flatness condition,
dJ J J = 0,
the connection
A(x) = J
(0)

x 2 + 1 (2)
2x (2)
J 2
+ 2
J +
x 1
x 1
x + 1 (1)
J +
x 1
x 1 (3)
J
x +1
(12)
is flat for any complex number x [3]. This is the crucial observation which indicates the model
to be (at least classically) integrable. Indeed, we can define the monodromy matrix

(x) = P exp A(x),
(13)
where is any path starting and ending at some point (, ) and wrapping the worldsheet cylinder once. Since the flatness of the connection ensures path independence we can choose to be
the constant path. Moreover, placing this loop at some other value of just amounts to a similarity transformation of the monodromy matrix. Thus we conclude that the eigenvalues of (x)
are time independent. Since they depend on a generic complex number x, we have obtained in
this way an infinite number of conserved charges thus assuring integrability. Finally we notice
that under periodic SP(2, 2) SP(4) gauge transformations the monodromy matrix transforms
by a simple similarity transformation so that the eigenvalues are also gauge invariant.
In the rest of this section we shall review the results of [4] and analyze the analytical properties
the quasi-momenta p and p defined from the eigenvalues
i p i p i p i p i p i p i p i p
e 1 , e 2 , e 3 , e 4 |e 1 , e 2 , e 3 , e 4
of the monodromy matrix (x). The eigenvalues are the roots of the characteristic polynomial
equation and thus they define an 8-sheet Riemann surface. These sheets are connected by several
cuts2 see Fig. 2whose branchpoints are the loci where the eigenvalues of the monodromy
matrix become equal. The quasi-momenta can jump by a multiple of 2 at points connected by
2 This connection appears as square root cuts, if the cut relates ps
(of ps)
among themselves, or by poles, if the pole
is shared between some p i and p j [4].
181
a cut.3 For example, for a cut going from the first to the second sheet, we will have
p 1+ p 2 = 2n,
x Cn12 ,
where p stands for the value of the quasi-momenta immediately above/below the cut. This
integer n, together with the filling fraction we shall introduce in next section, label each of the
cuts. Generically, we can summarize all equations as
pi+ pj = 2nij ,
ij
x Cn ,
(14)
where the indices i and j take values

2,
1,
2,
i = 1,
4,
3,
4
j = 3,
(15)
and we denote
p1,
3,
4 p 1,2,3,4 ,
2,
p1,
2,
3,
4 p 1,2,3,4 .
(16)
In Section 4 we will explain why we should consider only these 16 possible excitations. Moreover
to each cut we also associate the filling fraction

1
1 2 pi (x) dx,
Sij = 2
(17)
8 i
x
Cij
obtained by integrating the quasi-momenta around the square root cut. As before, the indices
2 and the minus sign for the remainrun over (15) and we should chose the plus sign for i = 1,
2.
Let us explain why we chose to integrate the quasi-momenta p(x)
ing excitations with i = 1,
around the cut with the seemingly mysterious 1 1/x 2 weight. In terms of the Zhukovsky variable z = x + 1/x Eq. (17) this expression takes the standard form (3). It was pointed out in [4,15]
and shown in [16] that these filling fractions are indeed the action variables of the theory. Indeed,
from the AdS/CFT correspondence these filling fractions are expected to be integers since they
correspond to an integer number of Bethe roots on the SYM side [7,17]. Indeed, the likely existence of the Bethe ansatz description [5,6] of the AdS5 S 5 superstring also implies this pole
structure of the exact quasi-momentum in a semiclassical limit. Moreover, in [9], where the S 5
subsector was studied it was clear that the quasi-momenta p(z) coming from the quantum Bethe
ansatz equations appears in the usual form, with unit residue for each pole, for the Zhukovsky
variable z. Thus (17) is the good starting point for the string quasi-classical quantization.
From (6), (12) it follows that
E 0
,
C 1 (x)C = ST (1/x), C =
0 E
which translates into the inversion symmetry
p 1,2 (x) = 2m p 2,1 (1/x),
p 3,4 (x) = +2m p 4,3 (1/x),
p 1,2,3,4 (x) = p 2,1,4,3 (1/x),
(18)
for the quasi-momenta.4

3 Note that the derivative of the quasi-momenta is a single valued function on the Riemann surface while p(x) is not.
4 Note that for p there is no 2 m imposed by requiring absence of time windings [4,18].
182
The singularities of the connection at x = 1 result in simple poles for the quasi-momenta.
These singularities come from the current J (2) in (12). This current is supertraceless because
it belongs to psu(2, 2|4) and so is its square due to the Virasoro constraints following from
the variation of the action with respect to the worldsheet metric. Together with the inversion
symmetry this forces the various residues to organize as follows
{p 1 , p 2 , p 3 , p 4 |p 1 , p 2 , p 3 , p 4 }
{ , , , | , , , }
.
x1
(19)
Thus we see that the residues at these points are synchronized and must be the same for the S 5
and the AdS5 quasi-momenta p i and p i . This is the crucial role of the Virasoro constraints which
will be of utmost importance in the remaining of the article.
Finally, for large x, one has

2
A g 1 + k g,
x
where k, defined bellow (9), is the Noether current associated with the left global symmetry.
Thus, from the behavior at infinity we can read the conserved global charges5 [19]
p 1
+E S1 + S2
p 2
+E + S1 S2
p
E S S
1
2
3
2
4
E + S1 + S2
(20)

.
p 1 x +J1 + J2 J3

p 2
+J1 J2 + J3

p 3
J1 + J2 + J3
p 4
J1 J2 J3
The finite gap method allow us to build, at least implicitly, classical solutions of the non-linear
equations of motion from the analytical properties of the quasi-momenta.6
As we shall see in the next section, the algebraic curve can also be turned into a powerful tool
to study the quantum spectrum, i.e., the energy level spacing, for energies close to that of a given
classical string solution.
3. Circular string solutions
In this section we will write down an important class of rigid circular strings studied in [12].
As we explain below they are particularly simple from the algebraic curve point of view and will
therefore provide us an excellent playground to check our method for some simple choice of
parameters. In terms of the AdS5 and S 5 embedding
coordinates,
we can represent this general
class of strings solutions with global charges E = E, J1 = J1 , . . . , as [12]

J3 i(w3 +m3 )
S2 i(w2 +k2 )
e
,
v2 + iv1 =
e
,
u2 + iu1 =
w3
w2
5 These are the bosonic charges, the ones which are present for a classical solution. Latter we shall consider all kind of
fluctuations, including the fermionic ones. Then we shall slightly generalize this expression to (28).
6 For the inverse problem of recovering the solutions from the algebraic curve see the monographs [20,21] for the
general formalism and [16] where this is carried over in the context of string theory for the classical bosonic string in
R S 3 AdS5 S 5 described by the KMMZ algebraic curve [7].

u4 + iu3 =

u6 + iu5 =
J2 i(w2 +m2 )
e
,
w2
v4 + iv3 =
J1 i(w1 +m1 )
e
,
w1
v6 + iv5 =
183
S1 i(w1 +k1 )
e
,
w1
E i
e ,
(21)
where the equations of motion, Virasoro constraints and (11) impose

1=
3

Ji
i=1
wi
E Sj
,
wj
2
1=
w2j = 2 + kj2 ,
j =1
2 =
2
Sj
j =1
wi2 = 2 + m2i ,
0=
3

i=1
Ji
2kj2
wj
2
kj Sj +
j =1
3

i=1
wi2 m2i
.
wi
Ji
3
mi Ji ,
i=1
wi2 + m2i
,
wi
(22)
As explained in Appendix A, for this family of solutions the representative g can be written as
g = e + g0 ,
where , are linear combinations of Cartan generators and g0 is a constant matrix. Then we
see that the current
J = g 1 dg,
and therefore also the flat connection A(x) in (12), are constant matrices! Then the computation
of the path order exponential (13) is trivial and the quasi-momenta p(x) are simply obtained
from the eigenvalues of 2
i A(x). For a detailed account see Appendix A.
4. Frequencies from the algebraic curve
In this section we will consider the quasi-classical quantization of the AdS5 S 5 superstring
in the language of the algebraic curve. As an example we will find the low lying energy spectrum
for the excitations around some simple classical string solutions.
As we have already mentioned in the introduction and in Section 2.2 the exact quasi-momenta
is made out of a large collection of poles. From (17) we infer the residue of each pole,
p
Sn

(xk )
+ ,
x xk
a=k
with
4 x 2
.
(x) = 2
x 1
(23)
These poles may then condense into square root cuts forming a classical Riemann surface like in
Fig. 2. The filling fraction and mode number of the cuts are in strict analogy with the amplitude
and mode number of a fourier mode in the usual flat space string. Then, to consider the quantum
fluctuations around this classical solution, amounts to adding small cuts, i.e., poles, to this curve.
184
Fig. 3. Some configuration of poles on the algebraic curve corresponding to the S 5 excitations (red) and AdS5 excitations
(blue). Black line denotes poles at 1, connecting 4 sheets with equal residues. The crosses correspond to the residue
+(x), while circles to residue (x). Physical domain of the surface lies outside the unit circle. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)
The key ingredient allowing us to do so is the knowledge of the residue (23) just like in the
example (2) in the introduction. The several possible choices of sheets to be connected by these
poles correspond to the several possible polarizations of the superstring, i.e., to the different
quantum numbers. The 16 physical excitations are the 4 + 4 modes in AdS5 and S 5 (Fig. 3) plus
the 8 fermionic fluctuations (Fig. 4).
Let us give a bit more of flavor to the discussion above. As we mentioned in the introduction,
the equations describing the eight sheet quasi-momenta can be discretized [5] yielding a set of
Bethe ansatz equations for the roots xi making up the cuts. The resulting equations resemble (4)
with an extra 2ni in the left-hand side
1
= 2ni + V (xi ).
x i xj
j =i
This means that we can think of xi as being the position of a particle interacting with many
other particles via a two-dimensional Coulomb interaction, placed in an external potential7 and
feeling an external force 2ni . What we are doing is, then, first considering a large number of
particles which will condense in some disjoint supports, the cuts, with each cut being made out
of particles with the same mode number ni . Then we add an extra particle with some other mode
number n. At the leading order, two things happen. The particle will seek its equilibrium point
in this background and will backreact, shifting this background slightly by its presence [8]. The
(AdS global time) energy E of the new configuration is then shifted. When adding N particles
we get precisely the quantum steps in the spectrum, i.e., (5).
Technically the computations can be divided into two main steps. In what follows we will use
the notation (16) intensively. We must solve (14) for all cuts of the Riemann surface where we
7 In (4) the potential is a quadratic one, for the actual Bethe equations it is something else.
185
Fig. 4. Some configuration of poles on the algebraic curve corresponding to the 8 fermionic excitations. Black line
denotes poles at 1, connecting 4 sheets with equal residues. The crosses correspond to the residue (x), while circles
to residue (x). Physical domain of the surface lies outside the unit circle.
now have p(x) p(x) + p(x) where p(x) is the quasi-momenta associated with the classical
solution.
When applied to the microscopic cut, i.e. pole, Eq. (14) gives us, to leading order, the position
ij
xn of the pole,
ij
ij
ij
pi xn pj xn = 2n, xn > 1,
(24)
2,
3,
4,
1,
2,
3,
4 and indicate which two sheets share the
where i < j are taking values 1,
pole. We refer to domain |x| > 1 as physical domain. The interior of the unit circle is just the
mirror image of the physical domain, as we saw in the previous section (18).
Then, to find p, and in particular the energy shift E, we must solve the same equations but
now in the macroscopic cuts
pi+ pj = 0,
ij
x Cn .
(25)
This linear problem is to be supplemented with the known analytical properties of p(x)
namely the asymptotic behavior presented below and the simple pole singularities with
residues (23). In this way we are computing the backreaction described above.
ij
Before proceeding it is useful to introduce some simple notation. We shall consider Nn excitations with mode number n between sheet pi and pj such that
ij
Nij
Nn
n
is the total number of poles connecting these two sheets. Moreover, each excitation has their own
quantum numbers according to the global symmetry. The S 5 , AdS5 and fermionic excitations can
186
then be identified as the several possible choices of sheets to be connected, see Figs. 3 and 4,
S5,
AdS5 ,
Fermions,
3),
(1,
4),
(2,
3),
(2,
4),
(i, j ) = (1,
3),
(1,
4),
(2,
3),
(2,
4),
(i, j ) = (1,
3),
(1,
4),
(2,
3),
(2,
4),
(i, j ) = (1,
3),
(1,
4),
(2,
3),
(2,
4).
(1,
(26)
The 16 physical degrees of freedom of the superstring are precisely these 16 elementary excitations, also called momentum carrying excitations [4,17].
When adding extra poles to the classical solutions its energy will be shifted by

1
Nij +
Nij ,
E = +
(27)
2
5
AdS
Ferm
where we isolated the anomalous part of the energy shift from the trivial bare part. Then, it
is convenient to recast (20), for the excitations, as
p 1
+/2 + N1 4 + N1 3 + N1 3 + N1 4
+/2 + N2 3 + N2 4 + N2 4 + N2 3
p 2
p
/2 N2 3 N1 3 N1 3 N2 3
3
p
/2 N1 4 N2 4 N2 4 N1 4
4
4
.

(28)
p 1 x
N
14
13
13
14

p 2
N2 3 N2 4 N2 4 N2 3
p
+N + N + N + N
3
p 4
2 3
1 3
1 3
2 3
+N1 4 + N2 4 + N2 4 + N1 4
These filling fractions Nijn are not independent. Any algebraic curve must obey the Riemann
bilinear identity (see Eqs. (3.38) and (3.44) in [4]). Since this was already the case for the classical
solution around which we are expanding, the new filling fractions are constrained by
ij
n
Nn = 0,
(29)
n
All ij
which is nothing but the string level matching condition in the algebraic curve language.
It is also important to note that sign of the residues can be summarized by the following for ij ij
mula
ref p k = (i k j k ) xn Nn ,
ij
x=xn
ij ij
ref p k = (j k i k ) xn Nn ,
ij
(30)
x=xn
2,
3,
4,
1,
2,
3,
4,
as summarized in Figs. 3 and 4.
with k = 1, 2, 3, 4 and i < j taking values 1,
In the following sections we shall analyze the quantum fluctuations around some simple classical solutions belonging to the family of rigid circular strings (21). We will do it in three main
steps. First we compute the quasi-momenta8 associated to each classical solution as explained in
Section 3 and in greater detail in Appendix A. Then we shall consider the fluctuations around the
8 Due to the simplicity of these solutions we could have computed the quasi-momenta by an alternative method, namely
using just the analytical properties presented in Section 2.2. This was done for the su(2) and sl(2) circular solutions in
[7] and [18].
187
classical solution which appear as new poles in the quasi-momenta. As explained above, we start
by finding the position of these new roots using (24) and then we shall compute the perturbation
p of the quasi-momenta by using, again, the analytical properties described in Section 2.2 plus
the knowledge of the poles positions found in the second step.
We can already notice that, using this procedure, one relies uniquely on considerations of
analyticity and needs not introduce any particular parametrization of the group element g(, )
for the fluctuations around the classical solution, contrary to what is usually done in this type
of analysis [1013]. It is also nice to see that the fermionic and bosonic frequencies appear, in
our approach, on a completely equal footing, both corresponding to simple poles which differ
only by the sheets they unitesee Figs. 3 and 4. Finally, in principle, we can apply our method
to any classical solution whereas the same generalization seems to be highly non-trivial to do
directly in the action since we no longer have a simple field redefinition to make it time and
space independent as was the case in [11,13].
4.1. The BMN string
We shall consider the simplest possible solution amongst the family of circular strings presented in Section 3, the rotating point like BMN string [22] moving around a big circle of S 5 .
For this solution all spins except for
J1 = J
are set to zero. Then we have m1 = 0, w1 = J , E = = J . For this solution the connection
A(x) presented in Appendix A is not only constant but also diagonal so we immediately find
p 1,2 = p 3,4 = p 1,2 = p 3,4 =
2J x
.
x2 1
(31)
We see that this is indeed the simplest 8 sheet algebraic curve we could have builtit has neither
poles nor cuts connecting its sheets other than the trivial ones at x = 1 (19).
We shall now study the quantum fluctuations around this solution. For the sake of clarity we
shall not write explicitly many of the quantities computed in the intermediate stepsthey can be
found in Appendix B.
To consider the 16 types of physical excitations we add all types of poles on the Figs. 3 and 4.
From (24) we find that the poles in the physical domain with |x| > 1, for this simple case, are all
located at the same position
ij
x n = xn =

1
J + J 2 + n2 .
n
Now we must find the quasi-momenta p(x) + p(x)
with poles located at (32) with residues (30) connecting the several sheets,
obeying the x 1/x symmetry property (18),
with residues 1 grouped as in (19),
with large x behavior given by (28).
(32)
188
From the requirements listed above one can easily write the expression for the quasi-momenta.
For example
p 2 = a +
+
+
x 1 x +1
p 3 = b +
x 1 x +1

4,
3,
4
i=3,

2,
1,
2
i=1,
(x 2i
)N 2i
n
xn2i
(x 3i
)N 3i
n
xn3i

4,
3,
4
i=3,

2,
1,
2
i=1,
(x 1i
)N 1i
n
1/x xn1i
(x 4i
)N 4i
n
1/x xn4i
(33)
(34)
where a,
b and , are constants to be fixed and the last terms ensure the right poles in
physical domain for p 1,4 (x) = p 2,3 (1/x). Similar expressions can be immediately written
down for p 2,3 with the introduction of two new constants a and b.
At this point we are left with the problem of fixing the eight constants
a,
+ , , + , .
a,
b,
b,
This is precisely the number of conditions one obtains by imposing the 1/x behavior at large x
for the quasi-momenta (28). The asymptotic of p 2 , p 3 , p 2 , p 3 fix the first four constants while
the remaining four equations, solvable only if the level matching condition (29) is satisfied, fix
the remaining coefficients and yield

n2 + J 2 J ij
1 ij
N ij +
N ,
Nn +
E =
(35)
J
2
5
n
All
AdS
Ferm
where we indeed recognize the famous BMN frequencies [22] in the anomalous part of the energy
shift.
4.2. The circular string in S 3 , the 1 cut su(2) solution
The next less trivial example is the simple su(2) rigid circular string [10]. Still it is simple
enough so that all results are explicit. This solution is obtained from the family of circular strings
in Section 3 by setting
m1 = m2 = m,
J1 = J2 = J ,
with all other spins set to zero. For this solution

E = = J 2 + m2 .
The quasi-momenta can be computed as explained in Appendix A. The AdS5 quasi-momenta are
obtained as for the BMN string
p 1,2 = p 3,4 =
2x
,
x2 1
(36)
while for the S 5 components p i we find that this solution corresponds to 1 cut between p 2 and
p 3 with mode number k = 2m, given by [7]

+ x 2x1 K(1/x)
p 1

+ 2x K(x) m
p 2
,
x 1
K(x)
(37)
m2 x 2 + J 2 ,
= 2
x
p 3
K(x)
+
m
2
x 1
p 4
2x K(1/x)
x 1
189
where we assume that m > 0 and branch cut goes to the left of x = 1 so that
K(x) = mx + O(1/x),
K(x) = J + O(x),
K(1) = K(1) = > 0.

In the rest of this section we will compute the quantum spectrum of the low lying excitations
around this solution. For simplicity we will consider the AdS5 , S 5 and fermionic fluctuations
independently assuming the level matching condition (29) to be satisfied for each of the sectors
separately. The result we give, however, is valid under the softer constraint (29) for all sectors,
as one can easily check.
4.2.1. Method of computation
Suppose we want to compute the variation of the quasi-momenta p(x) when a small pole is
added to some general finite gap solution with some square root cuts.
branch points
Since the
will be slightly displaced we conclude that p(x) behaves like x0 x x0 1/ x x0 near
each such point.
We are dealing with a 1-cut finite gap solution. Then, for p 2 , we can assume the most general
analytical function with one branch cut, namely f (x) + g(x)/K(x) where f and g are some
rational functions and K(x) was defined in (37). To obtain p 3 it suffices to notice that (25) is
simply telling us that p 3 is the analytical continuation of p 2 trough the cut. The remaining
quasi-momentum p 1,4 can then be obtained from this ones by the inversion symmetry (18). We
conclude that
g(1/x)
f (1/x) K(1/x)
p 1
g(x)
p 2 f (x) + K(x)
(38)
.
=
g(x)
p 3
f (x) K(x)
p 4
g(1/x)
f (1/x) + K(1/x)
The only singularity of p 2 apart from the branch cut are eventual simple poles at 1 and xn
and so the same must be true for f (x) and g(x). Then, just like in the previous example, these
functions are uniquely fixed by the large x asymptotics (28) and by the residues at xn (30) of the
quasi-momenta.
Finally, since the AdS5 part of the quasi-momenta of the non-perturbed finite gap solution has
no branch cuts their variations p i have the same form (33), (34) as for simplest BMN string.
4.2.2. The AdS5 excitations
This part is the simplest. The excitations live in the empty AdS5 sheets where the only impact
of the S 5 classical solution comes through the Virasoro constraints, by the residues at 1 (19).
Thus the p are nonperturbed and p i are the same as in BMN case (33), (34) with only AdS5
filling fractions N s being nonzero. Indeed, comparing (31) and (36) we see that we can completely recycle the previous computation provided we replace J by in the expression (32) for
the poles position. This leads immediately to
E =

AdS5 n

J 2 + m2 + n2 ij

Nn .
J 2 + m2
(39)
190
4.2.3. The S 5 excitations

We must now analyze the shift in quasi-momenta due to the excitation of the algebraic curve
2 4,
2 3,
1 4).
Since the AdS quasi-momenta are trivial, with no cuts,
by the four type of poles (1 3,
we obtain for p the same kind of expression we had for the BMN string (31), that is
2E x
,
p 1,2 = p 3,4 =
x2 1
(40)
where the constant factor was fixed by the asymptotics (28)

2E 1
p 1,2 p 3,4
.
x
Due to the Virasoro constraints the poles at 1 in the AdS5 and S 5 sectors are synchronized (19)
so that we merely need to compute f (x) and g(x) from the large x asymptotics (28) and the
residue condition (30) and extract, from these two functions, the residues at 1. This is done in
Appendix C.1. Let us just provide a glimpse of reasoning involved. Since the difference
p 3 = f (x) g(x)/K(x)
must have a single pole at xn13 with residue (xn13 ) whereas the sum
p 2 = f (x) + g(x)/K(x)
must be analytical, we can, in this way, read the residues of both f and g at this point. This kind
reasoning should be carried over for all the other excitations and for the points x = 1 and leads
to the ansatz (C.1), (C.2) where the only 3 constants left to be found can be fixed by the large x
asymptotics of the quasi-momenta.
One can then read of the energy shift from the large x asymptotics of the quasi-momenta
E =

1 3
xn (m + n) J
Nn13 + Nn24
+ Nn14
K(xn13 )
nxn14 2J
2m + n
+ Nn23
xn23
(41)
in terms of the positions of the roots obtained from the original algebraic curve through (24).
4.2.4. Fermionic excitations
We have 8 fermionic excitations but since p 1 = p 2 = p 3 = p 4 and p 1,2 = p 4,3 we
2 3,
3 2,
4 2)
poles and possibly another result for the
will get the same result for the (1 3,
2 4,
3 1,
4 1)
excitations. We can repeat the same kind of calculations we did for the S 5 ex(1 4,
citations to fix completely the quasi-momentasee Appendix C.2. Then, from the asymptotics
(28) we get
=

m + n
xn1 3
n
1 4

nxn J
+ Nn14 + Nn24 + Nn31 + Nn41
Nn13 + Nn23 + Nn32 + Nn42
(42)
191
4.3. The circular string in AdS3 , the 1 cut sl(2) solution

In Section 4.2 we analyzed in detail a simple su(2) solution with a particular mode number
k = 2m. In Appendix D we repeat the analysis for the general sl(2) circular string [12,13] which
also corresponds to a 1-cut algebraic curve but this time with an arbitrary mode number k for the
cut [18]. This solution is again contained in the family of circular strings written in Section 3. It
corresponds to two non-zero spins
S1 = S,
J1 = J ,
with mode numbers m1 = m, k1 = k constrained by the level matching condition

Sk + J m = 0
and frequencies w1 = J and w1 = w fixed by

w 3 k 2 + m2 + J 2 w + 2kmJ = 0.
For this solution = w 2 k 2 and the energy can be found from

S
.
E = 1+
w
(43)
In Appendix D we present the quasi-momenta associated to this classical solution and compute the fluctuation frequencies as we did for the su(2) string. These results, together with the
ones for the su(2) circular string, are summarized and discussed in the following sections.
5. Results, interpretation and 1-loop shift
5.1. Results
In this section we list all our results and introduce the notation usually used in the literature.
In the next section we shall analyze them, compare them and draw some conclusions.
5.1.1. Simple su(2) circular string
In Section 4.2 we found the level spacings around the simple su(2) circular solution, that
is the fluctuation frequencies of the effective quadratic Lagrangian obtained by expanding the
MetsaevTseytlin action (8) around this classical solution. In [11] this computation was performed having in mind the stability analysis and computation of the one-loop shift. The various
frequencies and corresponding degeneracies and origin can be summarized in Table 1.9 Using
the notation introduced in this table we can replace the explicit expressions for the position of
the roots found from (24) and recast our results (39), (41), (42) as

S
S
S
E =
Nn13 + Nn24 n+m
J + Nn23 n+2m
+ Nn14 n+ 2J
n

Nn14
+ Nn24
+ Nn31

F
n J +
2

+ Nn41
9 By expanding the GS action without imposing the Virasoro conditions from the beginning, one obtains, apart from
the frequencies listed in the above table, some massless modes with = n [10]. This Virasoro modes can be seen from
the Bethe ansatz point of view [23] if an extra level of particles with rapidities is introduced [9].
192

Nn13
+ Nn23
+ Nn32

+ Nn42
F
n+m

Nn13 + Nn14 + Nn23 + Nn24 nA .
(44)
We notice the appearance of constant shifts and relabeling of the frequencies when compared to
those in Table 1. We shall discuss this point below.
5.1.2. General sl(2) circular string
The same analysis can be carried over for the sl(2) circular string. In [12,13] this frequencies
were computed and the result can be summarized in Table 2.10
In the notation of the above table, the results (D.12), (D.9), (D.5), (D.4) derived in Appendix D
can be put together as
E =

A
A+
Nn13 + Nn24 nA + Nn23 nk
+ w + Nn14 n+k
w

Nn13
+ Nn14
+ Nn41

+ Nn42
Nn23 + Nn24 + Nn31 + Nn32

n
1
F
F
n+m/2k/2
m/2k/2
+
2
F
nm/2k/2
F
m/2k/2
1
+
2

S
Nn13 + Nn14 + Nn23 + Nn24 n+m
J .
(45)
Table 1
Simple su(2) frequencies
Eigenmodes

2
2
4
2 2
2 2
2J + n 2 J + n J + m n
2
2
2
J + n m
2
2
J + n
2
J + n2 + m2
S5
Fermions
AdS5
Notation
S
n
nS
nF
nA
Table 2
General sl(2) frequencies
AdS5
Eigenmodes
Notation
4S 2 2
4E
2
2 2
2
(
n ) + w (w kn) = 0
2
2
n +
n + > n
nA

Fermions
S5
2
2
(n + w 2J )2 + 12 ( 2 + J 2 m2 )

J 2 + n2 m2
nF
nS
10 The results in this table are slightly simplified compared to those usually presented in the literature, especially the
fermionic frequencies.
193
5.2. Explanation of shifts

Let us first look at the su(2) result (44) and pick one if the frequencies, say the first one
S
n+m
J.
(46)
We find two kinds of shifts relatively to the frequencies listed in Table 1, namely the constant
shift J and the shift in the fourier mode n n + m. The same shifts we observe for the sl(2)
frequencies.
Let us understand the origin of this shifts. For that purpose consider a system of two harmonic
oscillators,

x12 + x22 2 2
x1 + x22 ,
2
2
and suppose that, instead of quantizing this system, we chose to quantize the system obtained by
rotating x1 , x2 with angular velocity J , i.e., we move to the y frame
Lx =
x1 + ix2 = (y1 + iy2 )ei J t .

Then, we obtain11
Hy = Hx + J Lz ,
where Lz is the usual angular momentum, so that
y
En1 ,n2 = + ( J )n1 + ( + J )n2 .

Thus for the radially symmetric wave function, for which n1 = n2 (and in particular for the
ground state energy), the constant shifts cancel and we obtain the same energies as for the first
system. That, in general, the two results are different is obvious since the energy depends on the
observer.
The constant shifts mentioned above have exactly this origin. In fact, when expanding the
MetsaevTseytlin string action around the classical su(2) circular string one obtains an effective
time and space dependent Lagrangian whose , dependence can be killed by a change of frame
X = R(, )Y,
where X are the (bosonic) components of the fluctuations and R is a time and space dependent
rotation matrixsee for instance expression (2.14) in [11].12 The same kind of field redefinitions
are also present for the fermion fields. The time dependence of the rotation matrix gives the
constant shifts as in the simple example we just considered while the space dependence in this
change of frame is responsible for the relabeling of the mode numbers.
To make contact with the algebraic curve let us return to the frequency (46) we picked as
illustration. It corresponds to a pole from sheet p 1 to p 3 (or from p 2 to p 4 ) whose position is
fixed by (24). The result in the rotated frame, nS , would correspond to a pole with mode number
n + m whose position is given by

p 1 xn13 p 3 xn13 = 2n + 2m.
11 In the y frame the Lagrangian takes the form 2L = y 2 + y 2 (2 J 2 )(x 2 + x 2 ) + 2J y y 2J y y .
y
1 2
1 2
1
2
1
2
12 The same is true for the sl(2) circular string. The authors have moved to a different frame through a time and space
dependent rotationsee for instance Eq. (4.11) in [13]and should, therefore, measure shifted energies.
194
When plugging the actual expressions (37) for p 1 and p 3 in this equation we see that the 2m
disappears and the equation looks simpler than (24). However, for several cut solutions there is
no such obvious choice of mode numbers (or field redefinition which kills the time dependence
in the Lagrangian).
5.3. 1-loop shift and prescription for labeling fluctuation frequencies
To compute the 1-loop shift to the classical energy of a given solution, according to [24], one
has to sum over all energies of the modes in the expansion around the classical configuration
8

N
8

1
B
F
i,n
i,n .
lim
E1-loop =
2 N
n=N
i=1
i=1
However, the right-hand side is hard to define rigorously. For n each frequency behaves
like
B
|n| ciB + diB ,
i,n
F
i,n
|n| ciF + diF ,
and thus the sum is sensible to the labeling of the frequencies. The seemingly innocent redefinition
B
B
i,n+k
,
i,n
i
F
F
i,n
i,n+l
,
(47)
i

with integer shifts constrained by i ki li = 0 to ensure the convergence of the sum, does
change the result

ki2 li2 + 2ciB ki 2ciF li .
E1-loop E1-loop +
(48)
In Appendix E we discuss in greater detail the effect of these shifts.

One way to compute the frequencies is to expand the MetsaevTseytlin action around some
classical solution. Generically, the resulting quadratic Lagrangian is time and space dependent.
To eliminate this dependence, when possible, a field redefinition is performed. However, there
are several ways to do the field redefinition to get a time and space independent action. Different
choices will give different sets of frequencies related by transformations like (47) and will therefore lead to different results. In Appendix E we analyze this kind of dangers by focusing on two
explicit examples. Thus we need a solid prescription for the labeling of the frequencies.
Suppose we were semiclassically quantizing around some classical string solution in flat
space. Then we would expect to find some fluctuation frequencies, the zero modes, corresponding to an overall translation of the string solution and which should, therefore, carry no energy at
all. Then the usual prescription is to take i,0 = 0.
The zero modes should also exist for a string in the AdS5 S 5 space with a large amount of
isometries. Indeed, let us take our results and denote the contribution at n = 0 in (44) and (45)
su(2)
sl(2)
and E0 . Then we find that they are equal and given by
by E0
E0 =

AdS5
Nij +
1
Nij .
2
(49)
Ferm
In other words, the contribution to the anomalous part of zero modes for our labeling is zero!
Thus the prescription we used seems to be the precise analogue of the flat space fourier modes
prescription.
195
Table 3
General su(2) frequencies
S5
Fermions
AdS5
Eigenmodes
Notation
4J
4J
(2 n2 )2 w 2 (w1 m1 n)2 w 1 (w2 m2 n)2 = 0
2
1

2
2
n +
w12 +m22 2
)2 + J1 w1 + J2 w2
2
(n
n2 + 2
n+ > n
nS
nF
nA
Moreover, by construction, we have a good BMN limit. That is, when in the limit of very
small cuts with m 0 we recover the result (35) without any unusual shifts.13
Then, from (44) and (45), we can write the 1-loop shifts for the 1-cut circular solutions14
su(2)
E1-loop
N

1
S
S
S
F
=
4nA + n+m
+ n+2m
+ n+ 4nF 4n+m
,
lim
2 N
n=N
sl(2)
E1-loop =
N

1
A+
A
S
F
F
2nA + n+k
+ nk
+ 4n+m
4n+
.
lim
mk 4
n m+k
2 N
2
2
n=N
5.4. General su(2) results

Another interesting solution contained in the family of circular solutions described in Section 3 is the generalization of the simple su(2) solution to the case of two non-equal spins J1,2
with two different mode numbers m1,2 . The fluctuation frequencies associated with this solution
can be listed in Table 3 [12].15
Now, armed with our prescription, we can write the one loop shift unambiguously. Imposing
Good BMN limit (35) for vanishing filling fractions,
Proper zero mode behavior with n = 0 frequencies having trivial anomalous, part (49),
For m1 = m2 = m we should retrieve the simple su(2) result (44),
we get (for m1 + m2 0)

S
S
Nn13 + Nn24 n+m
w1 + Nn23 n+m1m2
+ w2 w1
E =
1
n
1 3
S+

Nn + Nn14 + Nn23 + Nn24 nA
Nn14 n+m
w2 w1 +
1 +m2
13 This is not the case for the frequencies listed in Table 2 for instance. For example, from this expression, we find, for

the fermionic frequencies, nF (n + k/2)2 + J 2 . See also discussion in Appendix E.
14 As for the simple example of the harmonic oscillators in the previous section, the sum of all constant shifts appearing
in (44) and (45) cancel so that only the shifts in mode number lead to a change of the final result. The difference with
respect to the sum with no shifts can be obtained from (48) and is equal to m2 / in both cases.
15 The fermionic frequencies for the general circular string of Section 3 can be computed (we shall publish our findings
elsewhere). In particular, for the su(2) general circular string we find the results listed in Table 3.
196

Nn14
+ Nn24
+ Nn31

+ Nn41

Nn13
+ Nn23
+ Nn32

+ Nn42
F m1 +m2
n+ 2
m1 +m2
F m1 m2
n 2
+
2
F m1 m2
2
+
.
2
6. Conclusions
We explained how to compute the quantum fluctuations around any classical superstring motion in AdS5 S 5 . These excitations include the fermionic, AdS5 and S 5 modes. We showed that
each mode correspond to adding a pole to a specific pair of sheets i, j of the algebraic curve. The
position of the pole is determined from the equation
ij
ij
pi xn pj xn = 2n,
and thus provides one with an unambiguous labeling for the frequencies. In particular we observe
the nice feature that for n = 0 this equation prescribes the pole at infinity, that is we find the
expected zero modes associated with global transformations under the isometries of the target
superspace.
Technically we computed the change in quasi-momenta due to the addition of these new poles
and read, from the large x asymptotics, the global charge corresponding to the AdS Energy, that
is the frequencies. However, since we computed explicitly the perturbed quasi-momenta we have
obtained not only the energy shift but actually all conserved charges!
Finally our method gives a new insight into the 1-loop shift analysis. Having at hand a simple
and universal framework to compute fluctuation frequencies, we can now prove general statements about the 1-loop shift of any classical string solution [25].
Acknowledgements
We would like to thank J. Penedones, P. Ribeiro, K. Sakai, A. Tseytlin, D. Volin, K. Zarembo
and especially V. Kazakov for many useful discussions. The work of N.G. was partially supported by French Government PhD fellowship, by RSGSS-1124.2003.2 and by RFFI project
grant 06-02-16786. P.V. is funded by the Fundao para a Cincia e Tecnologia fellowship
SFRH/BD/17959/2004/0WA9.
Appendix A. Quasi-momenta for a generic rigid circular string
To establish the link between the embedding coordinates solution (21) with the cosets notations we introduce the matrices
R=
3

e 2 (wi +mi )i R0 SU(4)
i=1
and
i
Q = e 2 1
2

i=1
e 2 (wi +ki )i+1 Q0 SU(2, 2),
197
where i are the Cartan generators,

1 = diag(+, +, , ),
and R0 = e42 e64 and Q0 = e
2 = diag(+, , +, ),

42
3 = diag(, +, +, ),

64
e
are constant matrices with

J1
J2
J3
(cos , sin cos , sin sin ) =
,
,
,
w1
w2
w3

E
S1
S2
,
,
(cosh , sinh cos , sinh sin ) =
w1
w2
, given, respectively by
and 42 , 64 , 42
64
0 1 0 0
0 0 0
1 1 0 0 0
0 0 1
0 1 0
2 0 0 0 1
0 0 1 0
0 0 0
0 1 0 0
0 0
1 0 0
1 1 0
0 0
,
0 1
2 0 0
2 0 1
0 0 1 0
1 0
0
0
,
0
0
0 1
1 0
.
0 0
0 0
Let us ignore for a moment the fact that, for a generic choice of mode numbers mi , kj , these
matrices are not always periodic. Then, to describe the circular solutions we can use, as representative g PSU(2, 2|4), the block diagonal matrix
Q 0
,
g=
(A.1)
0 R
which indeed leads to (21) under the map (10).
What is particular about this solution is that, as follows trivially from the form of the matrices
R and Q, the current
J = g 1 dg,
and therefore also the flat connection A(x) in (12), are constant matrices! Then the computation of (13) is trivial and the quasi-momenta p(x) are simply obtained from the eigenvalues of
2
i A(x).
Before going on let us comment on the subtle point ignored abovethe periodicity of the
rotation matrices R (and Q). For some integers mi we see that this matrix could become antiperiodic. This means that in principle we should use another representative, Rperiodic , for which
we should still have (10) but which should be periodic. However, if both R and Rperiodic obey
these equations this means that they are related by an anti-periodic SP(4) gauge transformation.
This means that for the purpose of computing the quasi-momenta p(x) we can indeed always
use the element (A.1) provided we keep in mind that if R is anti-periodic we can recover the real
quasi-momenta through
i p i p i p i p i p i p i p i p
e 1 , e 2 , e 3 , e 4 |e 1 , e 2 , e 3 , e 4 For the true representative Rperiodic

= ei p1 , ei p2 , ei p3 , ei p4 | ei p1 , ei p2 , ei p3 , ei p4 Using the anti-periodic R instead .
The same kind of statement hold for the AdS element Q.
198
The computation of the quasi-momenta is then straightforward. The S 5 components p i are

given in terms of the eigenvalues16 of the symmetric matrix
b+
c(1/x)
d(x)
a + (1/x)
a + (x)
d(1/x)
c(x)
b+
A(x)
=
(A.2)
,
b
c(1/x)
d(1/x)
a (x)
a (1/x)
d(x)
c(x)
b
with
m3 cos ,
a (x) = a(x)
m1 w1 x + (m2 w2 x) cos + x cos 2 (w1 + m1 x + (w2 m2 x) cos )
,
a(x)
=
x2 1
b = (m2 m3 ) cos sin ,
(m2 + m3 )x 2 (m2 m3 ) 2w3 x
sin sin ,
c(x)
=
x2 1
m1 + w1 x + (m2 w2 x) cos
d(x)
=
sin 2 ,
x2 1
while the AdS quasi-momenta p i are the eigenvalues of
b+
c(x)
d(x)
a + (1/x)
a + (x) d(x)
c(x)
b+
A(x)
=
(A.3)
,
b
c(x)
d(x)
a (x)
d(x)
c(x)
b
a (1/x)
with
2 k1 (x 2 1) cos
cosh + k2 cos ,
x2 1
b = (k2 cosh k1 ) sin ,
k2 (x 2 + 1) 2w2 x
sin sinh ,
c(x)
=
x2 1
k1 (x 2 + 1) 2w1 x
cos sinh .
d(x)
=
x2 1
For the simple su(2) or sl(2) solutions we have, amongst other conditions, = = 0 which
simplifies the computation drastically.
a (x) =
Appendix B. BMN string, details

This appendix serves as a complement to Section 4.1. The quasi-momenta with the correct
poles located at (32) and residues given by (30) is given by
p 2 = a +
+
+
x 1 x +1

4,
3,
4
i=3,
(x 2i
)N 2i
n
xn2i

4,
3,
4
i=3,
(x 1i
)N 1i
n
1/x xn1i
16 To each eigenvalues we might need to add a multiple of in such a way that its asymptotics become those prescribed
in Section 2.2. If R is periodic this multiple should contain an even number of s whereas if it is anti-periodic, we
should add n with n odd to each quasi-momentasee discussion in the text.
p 3 = b +
+
x 1 x +1
(x 3i
)N 3i
2,
1,
2
i=1,
xn3i
199
(x 4i
)N 4i
2,
1,
2
i=1,
1/x xn4i
where the last term guaranties that p 1,4 (x) = p 2,3 (1/x) have the right poles with the appropriate residues in the physical domain. Analogously
p 2 = a +
x 1 x +1
p 3 = b +
+
+
x 1 x +1
(x 2i
)N 2i
4,
3,
4
i=3,
xn2i
(x 3i
)N 3i
2,
1,
2
i=1,
3i
xn
(x 1i
)N 1i
4,
3,
4
i=3,
1/x xn1i
(x 4i
)N 4i
2,
1,
2
i=1,
1/x
n
4i
xn
(B.1)
(B.2)
and p 1,4 (x) = p 2,3 (1/x). From the large x behavior of these quasi-momenta one obtains
2n
2n
Nn1i ,
b = +
Nn4i ,
a =
J
J
n
n
2n
a = +
J
n
i=3,4,3,4
Nn1i ,
4,
3,
4
i=3,
the level matching condition (29) and

2n
+ =
J
n
+ =
2n
J
n
2n
b =
J
n
i=1,2,1,2
Nn4i ,
2,
1,
2
i=1,
ij
Nn ,
2
i=3,4,3,4 j =1,
ij
Nn .
4
4,
3,
4 j =3,
i=3,
Appendix C. SU(2) circular string, details

In this appendix we present the details of the calculations from Section 4.2 of the fluctuation
frequencies around the 1-cut SU(2) solution.
C.1. S 5 modes
2 4,
2 3,
1 4).
Thus
We start from the ansatz (38). We have four types of poles (1 3,

1
f (x) = p 2 (x) + p 3 (x)
2
must have simple poles at xn24 , xn13 with residues 12 (xn13 ) and 12 (xn24 ) respectively (see Fig. 3
or (30)). The same holds for

1
f (1/x) = p 4 (x) + p 1 (x) .
2
Moreover, since the residues of p i are connected to the AdS quasi-momenta (19) and these are
given by (40) we conclude that f (x) should be regular at x = 1. Thus we obtain

N 2 4 (x 2 4 )
(xn24 )
n
n
4 1 3)
.
+
(
2
f (x) =
(C.1)
2 x xn2 4
xn2 4 (1 xxn2 4 )
n
200
Then
g(x) =

K(x)
p 2 (x) p 3 (x)
2
must have simple poles at xn24 , xn13 and xn23 with residues 12 (xn13 ), 12 (xn24 ) and (xn23 )
respectively while
g(1/x) =

K(x)
p 4 (x) p 1 (x)
2
must have simple poles at xn24 , xn13 and xn14 with residues 12 (xn13 ), 12 (xn24 ) and (xn14 ) respectively. Contrary to f (x), this function may have poles at 1 so we arrive at
2 3
2 3
(x 1 4 )K(1/x 1 4 )
x+
n
2 3 (xn )K(xn )
g(x) = a + 2
+ 2
+
Nn14 n
N
n
x 1 x 1
xn14 (1 xxn1 4 )
x xn2 3
n

N 2 4 (x 2 4 )K(1/x 2 4 ) (x 2 4 )K(x 2 4 )
n
n
n
n
n
+
(C.2)
+ (24 13) .
2
xn2 4 (1 xxn2 4 )
x xn2 4
n
Finally the remaining constants are fixed by the large x asymptotic (28) to be
2 1 3

m Nn + Nn24 + 2mNn23 ,
a =
n

2 1 3

+ =
Nn + Nn24 xn13 (m + n) J K xn13
n

2m + n
+ Nn14 xn14 n 2J + Nn23
,
xn2 3
2 1 3

=
Nn + Nn14 + Nn23 + Nn24 n.
n
Then, from the residue at x = 1 we read E =
m2 +J 2
C.2. Fermionic modes

Arguments similar to the ones in the previous section lead to

2 x Nn14 + Nn24 Nn31 Nn41
Nn32 + Nn42 Nn13 Nn23
f (x) = 2
+x
x 1 n
xxn1 4 1
x xn3 2
and
x+
+
x2 1 x2 1

2 ( m2 + J 2 xn + n(1 xn2 ))(Nn14 + Nn24 + Nn31 + Nn41 )

+
(1 xxn )(xn2 1)
n

( m2 + J 2 xn + (n + m)(1 xn2 ))(Nn32 + Nn42 + Nn13 + Nn23 )
xn
,
(x xn )(xn2 1)
g(x) = b +
201
where
2m 1 3

Nn + Nn23 + Nn13 + Nn13 ,
b=
n

2 xn m2 + J 2
=
n Nn13 + Nn14 + Nn23 + Nn24
21
x
n
n
31
32

+ Nn + Nn + Nn41 + Nn42 ,

2
xn2 m2 + J 2 1 4

+ =
J nxn +
Nn + Nn24 + Nn31 + Nn41
2
xn 1
n

2

m + J 2 m + n 1 3
2 3
3 2
4 2
.
+
N
+
N
+
N
+
N
n
n
n
n
xn
xn2 1
The AdS5 part of the quasi-momenta is given by

2 x
Nn23 + Nn24
Nn13 + Nn14
p 2 (x) = 2
2x
,
+ 2
xxn 1
xn x
x 1

2 x
Nn31 + Nn32
Nn41 + Nn42
+ 2x
p 3 (x) = 2
+ 2
xxn 1
xn x
x 1
and p 1,4 (x) = p 2,3 (1/x). The constant can be found from fixing the residues at 1 for
p i and p i to be equal (19) and is given in the main text (42).
Appendix D. SL(2) circular string
The eigenvalues of (A.2) and (A.3) for the sl(2) circular string described in the beginning
of Section 4.3 yield the most general 1-cut quasi-momentum connecting p 2 and p 3 (due to the
x 1/x symmetry, p 1 and p 4 will be also connected by a cut, but this will be in unphysical
domain, that is inside the unit circle). Explicitly, we find
p 1,2 = p 3,4 = 2
and
Jx +m
x2 1
(D.1)
p 2 (1/x)
p 1
p 2 +p 2 (x)
,
=
p 3
p 2 (x)
p 4
+p 2 (1/x)
where [18]

(Cx + 1) x 2 2BCx + C 2
p 2 = k 1
C(x 2 1)

and
w

1
k
C+
,
2
C
B =1+
2S
.
w
(D.2)
(D.3)
202
From all solutions of (43), (D.3) for solutions for C and B we should pick the one for which we
have a real cut outside the unit circle.
In the rest of this appendix we will excite this solution by adding poles to quasi-momenta as
we did for the su(2) solution. In this way we shall find the energy shifts around this classical
solution. Moreover, as for the su(2) string, we shall consider the AdS5 , S 5 and fermions separately, assuming for simplicity the Riemann identity (29) to be satisfied for each of the sectors
separatelythe result, as before, holds if we relax this stronger assumption.
D.1. The AdS5 excitations
For these excitations the S 5 quasi-momenta remains untouched because its asymptotics do
not change and it is still only allowed to have simple poles at 1. Due to the Virasoro coupling
of these quasi-momenta to the AdS5 ones through the poles at 1 (19), we see that p i must
ij
have no poles at these points. The only poles of these quasi-momenta should be located at xn
with residues given by (30)see Fig. 3. Finally, as explained in Section 4.2.1, the perturbed
quasi-momenta should have inverse square behavior close to the branch points of the classical
solution.
Thus, from the same kind of reasoning we saw in the previous section for the su(2) circular
string, we find (D.6) for the AdS3 excitations connecting sheets (p 2 , p 3 ) and (p 1 , p 4 ) and (D.7)
for the remaining AdS5 excitations uniting (p 1 , p 3 ) and (p 2 , p 4 ). From the large x behavior (28)
of these quasi-momenta we read the energy shifts17

1 4
k n x 2 3 C 1 nw
nw
n
23
1 4 k + n xn C
E =
+
(D.4)
+ Nn
+
Nn
k xn2 3 + C 1
k
k xn1 4 + C
k
n
and

nw
+
E =
k
k(xn1 3 + C)
n

2 4
2 4
nw
2 4 K(xn )k + n(xn C)
+ Nn
,
+
k
k(xn2 4 + C)
Nn13
K(xn13 )k + n(xn13 C)
(D.5)
where, as for the su(2) string, we denote the square root in the classical solution (D.2) by K(x).
D.1.1. AdS3 excitationsdetails
In strict analogy to what we have already seen in for SU(2) solution, the two (left and right)
physical excitations inside the SL(2) sector described by the AdS3 -model, are given by the
poles, connecting the pairs of sheets (p 2 , p 3 ) and (p 1 , p 4 ). The number of such poles we denote
N2 3 and N1 4 respectively.
have no poles at 1 and must present
As explained above, the AdS5 excitations shifts of ps
an inverse square root behavior close to the branch points of the classical solution. Thus we can
write

xan
x a n
2
1 4
+
N
Nn23
,
p 2 (x) =
(D.6)
n
K(x) n
x xn2 3
x 1/xn1 4
17 These AdS excitations were also found in a similar way by K. Zarembo in relation with the finite size corrections
3
computation [30] (according to the private communication).
203
where
K(x)
x 2 2BCx + C 2
and the position of the roots is given by (24). Fixing the residues at xn23 and xn14 according to (30)
we have
an =
2Cxn23 (k n)
k(Cxn2 3
+ 1)
a n =
2C(k + n)
k(C + xn1 4 )
The large x asymptotic (28) is consistent if Riemann bilinear identity (29) is satisfied

Nn23 + Nn14 n = 0
and then the energy shift is given by (D.4).

D.1.2. The remaining AdS5 excitationsdetails
These correspond to simple poles connecting (p 1 , p 3 ) and (p 2 , p 4 ) for which
+cn x !
bn +cn x
1 3
)
Nn24 x(a n + bnK(x)
2 Nn x(an + K(x) )
p 2 (x) =
+
.
n (x xn13 )(x 1/xn13 ) (x xn24 )(x 1/xn24 )
(D.7)
Then p 3 , just as we saw for the su(2) solution, is the analytical continuation of p 2 through the
cut. In simpler terms, it corresponds to a simple change of sign of K(x) in the above expression.
Finally p 1,4 (x) = p 2,3 (1/x).
The undetermined coefficients are fixed by the residues

res p 1,3 = xn13 Nn13 ,
res p 2,4 = 0,
x=xn1 3
x=xn1 3
x=xn2 4
x=xn2 4

res p 2,4 = xn24 Nn24 ,
res p 1,3 = 0,
to be
an = 1,
bn = C
k(xn1 3 + C)
a n = 1,
2nxn13 + kK(xn13 )
cn =
2nxn24 + kK(xn24 )
,
bn = C
k(xn2 4 + C)
kK(xn13 ) 2Cn
k(C + xn1 3 )
cn =
kK(xn24 ) 2Cn
Hence, with the level matching condition

Nn13 + Nn24 n = 0,
we find, from the large x behavior, the energy shift (D.5).
k(C + xn2 4 )
204
D.2. The S 5 excitations

The S 5 quasi-momentum (D.1) has no branch cuts and thus p i will be of the same form as
we found for the BMN string except that the position of the roots, found from (24) is now given
by

J + J 2 + (n + m)2 m2
1 3
1 4
2 3
1 4
xn xn = xn = xn = xn =
(D.8)
n
instead of (32). The explicit expressions for p i are given in (D.X). This perturbation shifts the
residues at 1 from
(J m)
to some other values which we parameterize by

J + J eff m + meff .
The precise expressions for these shifts can be found in (D.10), (D.11). But then, since the AdS5
quasi-momenta only knows about the S 5 sector though the residues at these points, the perturbed
quasi-momenta p i + p i will be given by the same expression (D.2) with the trivial replacement
J , m J + J eff , m + meff .
The same is true for the energy, given in (D.3) so thatsee Appendix D.2 for detailswe can
immediately find

1 1 3

Nn + Nn14 + Nn23 + Nn24
(n + m)2 m2 + J 2 J .
E =
(D.9)
n
D.2.1. The S 5 excitationsdetails
As explained above the perturbed S 5 quasi-momenta are of the BMN form (B.1), (B.2)

4 x 2 Nn23 + Nn24 Nn13 + Nn14
+ 2
p 2 (x) = + 2
,
xn x
x xn x
x 1 n

4 x 2 Nn13 + Nn23 Nn24 + Nn14
+ 2
,
p 3 (x) = 2
(D.X)
xn x
x xn x
x 1 n
where xn is given by (D.8). Then, in the notation introduced above, the shift in the x = 1
residues is given by

1 3 + N 1 4 + N 2 3 + N 2 4 )(mn + J 2 J J 2 + n2 + 2mn)
(N
n
n
,
J eff = n n
(D.10)
n
(m2 J 2 )
while meff is given by
1 1 3

Nn + Nn14 + Nn23 + Nn24 n = 0,
J meff + J eff m =
n
due to the Riemann condition. Then, from (D.3), (43), we have
E =
w(k 2 + m2 + J 2 ) 3kmJ
w 3 kmJ
w
=
w,
w2
w2
(D.11)
205
where, using (43), (D.11), we have

w =
J eff
w 2 (m2 J 2 )
,
2
J w(k + m2 + J 2 ) 3kmJ
so that E will be given by (D.9).

D.3. Fermionic excitations
The fermionic excitations can be treated as for the su(2) string. As before we expect at
most two different answers for the energy shiftsone coming from the poles uniting the
1 4,
4 1,
4 2)
sheets another result for the (2 3,
2 4,
3 1,
3 2).
From the expressions in the next
(1 3,
section we find

Nn13 + Nn14 + Nn41 + Nn42 (1)
Nn23 + Nn24 + Nn31 + Nn32 (2)
=
n +
n ,
n
(D.12)
where
(1)
n =
(2)
n =
(2m + k) 2J C kC 2
k(C 2 1)(C 1 xn1 3
+ 1)
(2m k)C 2 2J C + k
k(C 2
1)(Cxn2 3
+ 1)
n(C 1 xn13 1)
k(C 1 xn1 3 + 1)
n(Cxn23 1)
+
k(Cxn2 3 + 1)
nw
,
k
nw
,
k
with the position of the fermionic poles being given by (24), in terms of the algebraic curve for
the classical solution.
D.3.1. Fermionic excitationsdetails
The S 5 part of the quasi-momenta is given by
p 2 (x) = +
p 3 (x) =
4x
(x 2 1)
4x
(x 2 1)
N 3 1 + N 4 1
n
xxn3 1 1
N 1 4 + N 2 4
n
xxn3 1 1
x xn3 1
Nn32 + Nn42
Nn13 + Nn23
x xn3 1

,

,
whereas the AdS5 part is more complicated. Parameterizing p as we did for the su(2) string in
(38), we have

x Nn23 + Nn24 Nn31 Nn32 Nn23 + Nn24 Nn31 Nn32
,
+
x
x2 1 n
x xn2 3
xxn2 3 1

2 3
x xK(xn23 )

g(x) = 2
+
a
x
+
b
Nn + Nn24 + Nn31 + Nn32
n
n
x 1 n
x xn23

1 3

1 3
xxn K(1/xn13 )
1 4
4 1
4 2
,
+ a n x + bn Nn + Nn + Nn + Nn
xxn1 3 1
f (x) =
206
where the remain constants are given by
an = C
a n = C
bn = C
(C 2 1)(k 2n)xn23 + 2Cm 2J

(C 2 1)(Cxn2 3 + 1)k
(C 2 1)k 2(m + n) + 2C(Cn + J )

(C 2 1)(C + xn1 3 )k
(C 2 1)k 2C 2 (m + n) + 2(n + CJ )
(C 2 1)(C + xn2 3 )k
(C 2 1)(k + 2n)xn13 + 2Cm 2C 2 J

.
bn = C
(C 2 1)(C + xn1 3 )k
Then the energy shifts can be read from the large x asymptotics and are given in (D.12).
Appendix E. Ambiguities due to shifts
To compute the 1-loop shift one must sum all frequencies. This sum, however, is sensitive to
the way the frequencies are labeled. Let us demonstrate this on a simple example. Consider the
sum
1
(2n n+m nm ),
2 n=
with n = n and assume that for large mode number, n |n| + . Naively this sum is zero
if m is integer, since all terms cancels among each other if we allow renumbering of the terms.
However a more careful analysis shows that this is not the case
N
1
(2n n+m nm ) =
2
n=N
N
(n nm ) = m2 + O(1/N).
n=Nm+1
Thus one should be very careful calculating 1-loop shift having a frequencies at hand because
ambiguities can easily arise. Consider, for example, the equation for the bosonic frequencies for
the general 3J solution [12]

J J J
P8 1 2 3 () = 2

2 4
3

2
Ji
2 2
4 n
(wj mj n)2
wi
i=j
+8
3

i=j =k=i
Ji
(wj mj n)2 (wk mk n)2 = 0.
wi
This solution can be smoothly deformed to general SU(2) solution for which J3 0 while
preserving all constraints (11). In this limit we find

2
2
su(2)
,
P8J1 J2 0 () = P4 () 2 n2 4 m3 n m23 + 2
su(2)
where the quartic polynomial P4 () is the one appearing in Table 3 and gives us the usual
SU(2) modes whereas the remaining equations yields the frequencies

(n + m3 )2 + 2 + w3 ,
(n m3 )2 + 2 w3
207
instead of the two n2 + 2 we read from Table 3. From the above explanation this ambiguity
converts into an extra contribution of m23 / to the 1-loop shift.
Moreover, we also found contradictory results in the literature. For the simple SU(2) solution,
in [11,26,27] the sum over fermionic frequencies nF is taken over the integers for even m and
over Z + 1/2 for odd m while in [28,29] the sum always goes over the integers. We found that
the fermions will indeed be summed over integer ns. The same kind of mismatch appears for
the sl(2) circular string. For example, in [13,2931] the fermionic frequencies nF are summed
F
F
with n integer whereas we found n+m/2k/2
and nm/2k/2
that is, the frequencies have
half-integer arguments if m + k is oddsee (45). In view of this discrepancies we also repeated
the calculation for the frequencies directly from the expansion of the string action using a coset
representative parameterized as in [14]. We also found the same kind of field redefinitions which
are trivially related to the R and Q matrices written in Section 3. For the simple SU(2) solution
the field redefinitions always leave the fermions periodic whereas for the sl(2) string they are
periodic (anti-periodic) for m + k even (odd) in agreement with the calculation presented in
this paper. Shifts changing integers into half-integers are no longer are related by the simple
expressions of the form m2 / like in the previous example. However, the sum over fermions can
be replaced by integral with exponential precision and therefore this shifts may end up being not
so harmful.
References
[1] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S 5 background, Nucl. Phys. B 533 (1998) 109,
hep-th/9805028.
[2] G. Mandal, N.V. Suryanarayana, S.R. Wadia, Aspects of semiclassical strings in AdS5 , Phys. Lett. B 543 (2002) 81,
hep-th/0206103.
[3] I. Bena, J. Polchinski, R. Roiban, Hidden symmetries of the AdS5 S 5 superstring, Phys. Rev. D 69 (2004) 046002,
hep-th/0305116.
[4] N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, The algebraic curve of classical superstrings on AdS5 S 5 ,
Commun. Math. Phys. 263 (2006) 659, hep-th/0502226.
[5] G. Arutyunov, S. Frolov, M. Staudacher, Bethe ansatz for quantum strings, Phys. Rev. D 66 (2002) 010001, hepth/0406256.
[6] N. Beisert, M. Staudacher, Long-range PSU(2, 24) Bethe ansaetze for gauge theory and strings, Nucl. Phys. B 727
(2005) 1, hep-th/0504190.
[7] V.A. Kazakov, A. Marshakov, J.A. Minahan, K. Zarembo, Classical/quantum integrability in AdS/CFT, JHEP 0405
(2004) 024, hep-th/0402207.
[8] N. Beisert, L. Freyhult, Fluctuations and energy shifts in the Bethe ansatz, Phys. Lett. B 622 (2005) 343, hepth/0506243.
[9] N. Gromov, V. Kazakov, K. Sakai, P. Vieira, Strings as multi-particle states of quantum sigma-models, Nucl. Phys.
B 764 (2007) 15, hep-th/0603043.
[10] S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS5 S 5 , Nucl. Phys. B 668 (2003) 77, hep-th/0304255.
[11] S. Frolov, A.A. Tseytlin, Quantizing three-spin string solution in AdS5 S 5 , JHEP 0307 (2003) 016, hepth/0306130.
[12] G. Arutyunov, J. Russo, A.A. Tseytlin, Spinning strings in AdS5 S 5 : New integrable system relations, Phys. Rev.
D 69 (2004) 086009, hep-th/0311004.
[13] I.Y. Park, A. Tirziu, A.A. Tseytlin, Spinning strings in AdS5 S 5 : One-loop correction to energy in SL(2) sector,
JHEP 0503 (2005) 013, hep-th/0501203.
[14] L.F. Alday, G. Arutyunov, S. Frolov, New integrable system of 2dim fermions from strings on AdS5 S 5 ,
JHEP 0601 (2006) 078, hep-th/0508140.
[15] N. Beisert, V.A. Kazakov, K. Sakai, Algebraic curve for the SO(6) sector of AdS/CFT, Commun. Math. Phys. 263
(2006) 611, hep-th/0410253.
[16] N. Dorey, B. Vicedo, On the dynamics of finite-gap solutions in classical string theory, JHEP 0607 (2006) 014,
hep-th/0601194.
208
[17] N. Beisert, V.A. Kazakov, K. Sakai, K. Zarembo, Complete spectrum of long operators in N = 4 SYM at one loop,
JHEP 0507 (2005) 030, hep-th/0503200.
[18] V.A. Kazakov, K. Zarembo, Classical/quantum integrability in non-compact sector of AdS/CFT, JHEP 0410 (2004)
060, hep-th/0410105.
[19] G. Arutyunov, S. Frolov, JHEP 0502 (2005) 059, hep-th/0411089.
[20] O. Babelon, D. Bernard, M. Talon, Introduction to Classical Integrable Systems, Cambridge Univ. Press, 2003.
[21] E.D. Belokolos, A.I. Bobenko, V.Z. Enolskii, A.R. Its, V.B. Matveev, AlgebroGeometric Approach to Nonlinear
Integrable Equations, Springer, Berlin, 1994.
[22] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-YangMills, AIP
Conf. Proc. 646 (2003) 3.
[23] N. Gromov, V. Kazakov, P. Vieira, Classical limit of quantum sigma-models from Bethe ansatz, hep-th/0703137.
[24] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002) 007,
hep-th/0204226.
[25] N. Gromov, P. Vieira, in preparation.
[26] S.A. Frolov, I.Y. Park, A.A. Tseytlin, On one-loop correction to energy of spinning strings in S5 , Phys. Rev. D 71
(2005) 026006, hep-th/0408187.
[27] S. Schafer-Nameki, Exact expressions for quantum corrections to spinning strings, Phys. Lett. B 639 (2006) 571,
hep-th/0602214.
[28] N. Beisert, A.A. Tseytlin, K. Zarembo, Matching quantum strings to quantum spins: One-loop vs. finite-size corrections, Nucl. Phys. B 715 (2005) 190, hep-th/0502173.
[29] R. Hernandez, E. Lopez, Quantum corrections to the string Bethe ansatz, JHEP 0607 (2006) 004, hep-th/0603204.
[30] S. Schafer-Nameki, M. Zamaklar, K. Zarembo, Quantum corrections to spinning strings in AdS5 S 5 and Bethe
ansatz: A comparative study, JHEP 0509 (2005) 051, hep-th/0507189.
[31] N. Beisert, A.A. Tseytlin, On quantum corrections to spinning strings and Bethe equations, Phys. Lett. B 629 (2005)
102, hep-th/0509084.
AdS/CFT correspondence as a consequence

of scale invariance
Hikaru Kawai a,b , Takao Suyama a,
a Department of Physics, Kyoto University, Kitashirakawa, Kyoto 606-8502, Japan
b Theoretical Physics Laboratory, The Institute of Physics and Chemical Research (RIKEN),
Wako, Saitama 351-0198, Japan

Abstract
We study an anisotropic scale transformation in the worldsheet description of D3-branes in type IIB
theory, and show that the transformation is really a symmetry in a region near D3-branes. AdS/CFT correspondence follows from this symmetry. We will explicitly show that Wilson loops in N = 4 supersymmetric
YangMills theory and minimal surfaces in AdS5 are related by the symmetry. The functional form of a supersymmetric Wilson loop operator is naturally derived from our worldsheet point of view.
1. Introduction
Since Maldacena proposed the conjectured relation, known as AdS/CFT correspondence [1],
between N = 4 supersymmetric YangMills theory (SYM) and type IIB string theory on AdS5
S 5 background, a vast amount of researches have been reported to provide pieces of supporting
evidence for the correspondence as many as possible. Many of such researches discuss some
limiting situations, for example, in the large N limit, where N is the rank of the gauge group.
On the expected range of validity, however, there does not seem to be a consensus. The most
well-studied limit is to take the limit of both the large N and the large t Hooft coupling .
E-mail addresses: hkawai@gauge.scphys.kyoto-u.ac.jp (H. Kawai), suyama@gauge.scphys.kyoto-u.ac.jp

(T. Suyama).
doi:10.1016/j.nuclphysb.2007.07.031
210
H. Kawai, T. Suyama / Nuclear Physics B 789 (2008) 209224
The correspondence claimed for this limit is sometimes referred to as the weakest version of
AdS/CFT correspondence.
One of the claims of the correspondence which is now well established is the correspondence
between Wilson loops W (C) in N = 4 SYM and minimal surfaces in the AdS5 background
[2,3]. The expectation value of W (C) is related to the partition function of a single type IIB string
whose boundary lies on C embedded in the boundary of AdS5 . A similar correspondence
was already anticipated in old days, and AdS/CFT correspondence provides a concrete example
which can be verified quantitatively.
In this paper, we would like to provide a natural point of view for understanding why such a
correspondence holds. It is based on a perturbative worldsheet description of D3-branes in the
flat spacetime. We consider an anisotropic scale transformation
x cx ,
y I c1 y I ,
(1.1)
where x is the longitudinal coordinates and y I are the transverse coordinates of the D3-branes.
For simplicity, we call this the scale transformation. We will show that the correspondence between the Wilson loops and the minimal surfaces follows from the invariance under the scale
transformation, which we call in this paper the scale invariance. In a region near D3-branes, this
invariance can be realized in the worldsheet description, as far as closed string loops are suppressed. The existence of the scale invariance immediately implies, for example, the weakest
version of the correspondence.
This paper is organized as follows. In Section 2, we provide an argument which indicates
the importance of the scale invariance in AdS/CFT correspondence. Section 3 is devoted to the
review of boundary states in the GreenSchwarz formalism [4]. Since the spacetime symmetries
play crucial roles, and also the RR states are important, it is convenient to employ the Green
Schwarz formalism. In Section 4, we define the scale transformation of the worldsheet variables,
and show that the free energy of type IIB string in the presence of D3-branes is invariant under the
transformation in a region near D3-branes. In our worldsheet description, the functional form of
a supersymmetric Wilson loop, which is studied in [5,6], is naturally derived without the detailed
knowledge of the supersymmetry transformation, as shown in Section 5. Section 6 is devoted to
discussion.
2. Scale invariance in AdS/CFT
Let us recall the metric of the solution of type IIB supergravity for N D3-branes
4
ds = 1 + 4
y
2
1
4
dx dx + 1 + 4
y
1
2
dy I dy I ,
(2.1)
where and run from 0 to 3, and I runs from 4 to 9. Here

= gs N
(2.2)
is the t Hooft coupling. We have defined y 2 = y I y I . The string length ls has chosen to be unity.
In a region y 4 , which we call the near-horizon region, the metric behaves as

dy 2
ds 2 4 y 2 dx dx + 2 + 4 d52 ,
(2.3)
y
211
Fig. 1. A worldsheet with four boundaries. Three of them are on the D3-branes. The remaining one of them, denoted
as C, is attached to a source, say another D-brane, apart from the D3-branes.
Fig. 2. A worldsheet near D3-branes. Open strings propagate inside the outer cylindrical worldsheet.
after the rescaling y 4y. This is the metric on the direct product of AdS5 , in the Poincar
coordinates, and S 5 . It is easily recognized that the scale transformation
x =
x ,
y I =
y I ,
(2.4)
is an isometry of the metric (2.3). This transformation shrinks the D3-branes in the longitudinal
directions x and expands in the transverse directions y I . This is actually an SO(1, 1) subgroup
of the isometry SO(4, 2) of AdS5 .
Let us examine the transformation property of the original metric (2.1) under (2.4). We obtain

y4
ds 2 =
(2.5)
ds 2 .
For a fixed , the variation of the metric under the transformation (2.4) is negligible in the nearhorizon region. One of the implications of the existence of the near-horizon region is as follows.
Consider an observable O[g] which depends only on the near-horizon region of the metric (2.1).
The scale transformation of O[g] is negligible in the near-horizon region, due to (2.5), and therefore O[g] is invariant there. In this sense, the scale invariance appears in the near-horizon region.
We would like to extend the above argument on classical gravity to string theory. We will
define in Section 4 the scale transformation (2.4) for worldsheet variables, and examine the invariance of the total system of strings and D3-branes in a similar sense discussed above. In the
rest of this section, we discuss implications of the existence of the symmetry in string theory.
To be concrete, let us consider a system of a string and N D3-branes, depicted in Fig. 1. The
string has a boundary C which is away from the D3-branes. One can imagine that there is another
D-brane on which C lies. The worldsheet of the string may have the other boundaries which are
on the D3-branes. We take C to be within the near-horizon region, in order for the system to have
the scale invariance.
On the one hand, we can expand the system in the x -directions and shrink in the y I directions by the scale transformation (2.4). Then, the worldsheet will have a configuration
depicted in Fig. 2. It is interesting to notice that this worldsheet configuration can be regarded
as a Feynman diagram for a Wilson loop W (CYM ) for a loop CYM in the double-line notation.
212
Fig. 3. A worldsheet far from D3-branes. The presence of the D3-branes are described by propagations of closed strings.
A worldsheet with more boundaries on the D3-branes corresponds to a more complicated Feynman diagram. The size of CYM can be taken to be as large as we want, compared with the string
scale. Therefore, the open strings propagating on the D3-branes have small momenta, and only
the contributions from the massless states of the open strings survive. These arguments imply that
the summation of the worldsheets with a fixed boundary CYM in the presence of the D3-brane
background actually evaluates a Wilson loop in N = 4 SYM in four dimensions. We discuss the
functional form of the Wilson loop in more details in Section 5.
On the other hand, we can perform another scale transformation which shrinks the system
in the x -directions and expands in the y I -directions, while keeping C within the near-horizon
region. Since this system is invariant under this transformation, the worldsheet path-integral for
this transformed system should provide another way of evaluating the Wilson loop discussed
above. We assume that the near-horizon region is large enough in order for the classical gravity
approximation to be valid, which is the case if we take a large enough . Since the worldsheet is
far away from the D3-branes in this case, the dominant contributions come from a worldsheet
which has only one boundary Cg related to C by the scale transformation. The presence of the
other boundaries lying on the D3-branes is now encoded by insertions on of vertex operators
of massless closed string states, which describe the exchange of massless states between the
D3-branes and . The summation over worldsheets with various numbers of boundaries then
corresponds to the summation of all possible massless state exchanges. As a result, feels that
the spacetime surrounding is curved, and therefore, the worldsheet path-integral is dominated
by a worldsheet with the boundary Cg which is the minimal surface in the near-horizon geometry
of the D3-branes. See Fig. 3.
We have argued that the scale invariance relates the Wilson loop in N = 4 SYM and the
minimal surface in AdS5 S 5 . It should be noted that we have shown an equivalence for the case
CYM = Cg . In fact, the length scales l of the loops are
l(Cg ) < l(C) < l(CYM ).
(2.6)
However, since the Wilson loop and the area of the minimal surface, both defined appropriately,
are independent of the length scale of the loop, the equivalence should also hold for the case
CYM = Cg which is the equivalence that AdS/CFT correspondence claims.
In summary, we have shown that the presence of the scale invariance implies AdS/CFT correspondence. The important point of our argument is that the scale invariance exists at any
intermediate situation, like the one in Fig. 1, not only at the two limiting cases (Figs. 2, 3). It
is also important that the quantities we have discussed are independent of the length scale, in
213
Fig. 4. A worldsheet with a handle which may be elongated beyond the limit of the near-horizon region.
order for the ordinary claim of AdS/CFT correspondence to hold. In Section 4, we will show that
the desired symmetry really exists if we concentrate on the dynamics in the near-horizon region.
As a result, the weakest version of AdS/CFT correspondence follows.
It would be important to mention the effects of string loops to the above argument. Throughout this paper, we will not discuss worldsheets with handles. In addition to the technical difficulty
of summing over string loop effects, there is a more serious issue. If we allow a handle on the
worldsheet, then the handle may be elongated, as depicted in Fig. 4. Since the handle includes the
propagation of massless closed string states, the worldsheet may feel the spacetime outside the
near-horizon region. This implies that we cannot concentrate our attention on the near-horizon
region appropriately, and the scale invariance is not available. In the following, we restrict ourselves to the standard large N limit, that is, we take
N ,
= gs N = fixed.
(2.7)
To show the scale invariance, we will use the boundary state [4] of the D3-brane in the Green
Schwarz formalism in the light-cone gauge, which we review in the next section.
3. Boundary state
In terms of closed strings, D-branes are described by boundary states. Since D-branes in type
IIB string preserve half of the 32 supercharges, the boundary state |B should satisfy

|B = 0,
Q + Q
(3.1)
where is a product of gamma matrices. For a D3-brane extending along x 0,1,2,3 -directions,
= 0 1 2 3
(3.2)
is taken. We often denote the preserved supercharges as Q . Note that these are the supercharges
of N = 4 SYM on the D3-branes.
It will be convenient to employ the GreenSchwarz formalism which makes the spacetime
supersymmetry manifest. In addition, it is rather easy to handle RR fields in this formalism.
For notations and conventions, we basically follow [7]. In the GreenSchwarz formalism, it is
convenient to take the light-cone gauge to quantize the theory, but the Lorentz invariance is not
manifest. In addition, the coordinate fields in the light-cone directions have to obey the Dirichlet
boundary condition [4], and therefore, the branes which are naturally realized as boundary states
214
are instanton-like objects. In the following, we review the construction of the boundary state for
a D3-brane [4].
For the boundary state of the D3-brane, the condition (3.1) becomes

a
b |B = 0,
Q + iM ab Q
(3.3)
a

b |B = 0,
Q + iM a b Q
(3.4)
where we have decomposed the SO(9, 1) spinors into SO(8) spinors, and we have defined
ab
a b

M ab = 1 2 3 4 ,
(3.5)
M a b = 1 2 3 4 .
There appears i in front of M ab and M a b in (3.3), (3.4) due to the double Wick rotation for
obtaining the Euclidean D3-brane from the physical one.
The D3-brane we consider is extended in x 1,2,3,4 -directions, and localized in the other directions including the light-cone directions. This situation is realized by the condition
i
j
n M ij n |B = 0 (n = 0),
(3.6)
where ni and ni are the oscillators of the bosonic coordinate fields X i with i = 1, . . . , 8, and

I44
0
ij
M =
(3.7)
.
0
I44
The condition for fermions S a and S a should be compatible with (3.6). That is,

j
Q , ni M ij n |B = 0
should be imposed. Recall that the explicit forms of the supercharges are
Qa = 2p + S0a ,

1
a
Sn
ni .
Qa =
i aa
p+
nZ
From (3.8), we obtain

a
b
|B = 0.
Sn + iM ab Sn
(3.8)
(3.9)
(3.10)
(3.11)
Note that the condition (3.11) for n = 0 is nothing but the BPS condition (3.3).
The conditions (3.6) and (3.11) determine the dependence of the boundary state on the nonzero modes as

1
j
ij i
ab a b
|B = exp
(3.12)
M n n iM Sn Sn |B0 .
n
n=1
The dependence of |B0 on the zero modes of S a and S a is determined as follows. The lowest
energy subspace of the Hilbert space of the left movers is spanned by the vector representation |i
and the conjugate spinor representation |a,
and similar for the right movers. |B0 is thus constructed from a sum of tensor products of them. Since |B is a source of the graviton, it is a
bosonic state. We assume

ij
1,2,3,4
5,6,7,8,+,

a b

|b p
=0 x
=0
|0n ,
|B0 = c |i|j + c |a
(3.13)
n=1
215
where |0n is the Fock vacuum of the nth mode. The condition (3.3) implies
ia a cij iM ab j bb ca b = 0,
ia a a b
(3.14)
ab j bb ij
+ iM
c = 0.
(3.15)
It is easy to check that

cij = CM ij ,
ca b = iCM a b ,
(3.16)
is a solution, where C is a constant. We obtain

p 1,2,3,4 = 0 x 5,6,7,8,+, = 0
|b
|0n .
|B0 = C M ij |i|j iM |a

a b
(3.17)
n=1
It is straightforward to show that this state also satisfies (3.4). The constant C is determined to
be unity by requiring the st duality.
4. Scale transformation in worldsheet theory
We consider a perturbative description of type IIB string with N D3-branes in the flat background. The worldsheet action is that of free fields,

1
1
S=
(4.1)
d 2 X i X i iS a S a i S a + S a ,
2
where we choose = 12 in this section. The presence of the D3-branes is encoded by including
contributions from worldsheets with boundaries. The back-reaction of the D3-branes to the background field configuration is taken into account by summing over all possible worldsheets with
various number of boundaries. It is obvious that the worldsheet action (4.1) is not invariant under
the scale transformation (2.4). Therefore, the scale invariance is not manifest in the worldsheet
description.
In this section, we define the scale transformation of the worldsheet degrees of freedom, in
terms of closed strings in a suitable manner, and first examine the transformation property of
the boundary state constructed in the previous section. Next, we show that the presence of the
D3-branes enables the scale transformation to be really a symmetry of the total system in the
near-horizon region.
4.1. Definition of scale transformation
We define the scale transformation of X i and their canonical momenta P i at = 0 as
X i ( ) =
M ij X j ( ),
(4.2)
P ( ) =
M P ( ).
(4.3)
ij
Note that we realize the transformation (2.4) as a canonical transformation. Recall that the mode
expansions of X i and P i at = 0 are

in
1 i
i
i
i
X ( ) = x + i
(4.4)
e
,
n
2
n n
n=0
216
2P i ( ) = p i +

n=0
in
i
ni + n
e
.
(4.5)
These imply the scale transformation of the oscillators as

j
ni =
M ij n ,
(4.6)
j
=
M ij n .
(4.7)
ni
For later use, it should be noted that the light-cone momentum p + also transforms as
p + =
p + ,
(4.8)
since the light-cone coordinates are taken to obey the Dirichlet condition.
The transformation of the fermionic oscillators are determined by using the transformation of
the supercharges
Q = Q ,
2
(4.9)
which is derived from the transformation (4.3) and the relation Q2 P . Note that this is equivalent to the commutation relation between the supercharges and the dilatation operator in N = 4
SYM. Evaluating
[Q , ni ] and
[Q , ni ] in two ways, we obtain
b
,
Sna = i
M ab Sn
a
ab b
Sn = i
M Sn ,
(4.10)
(4.11)
where we have used the transformation property (4.8). This transformation also preserves the
commutation relations. Note that the symmetric property of M ab is crucial for the preservation
of the commutation relations. The transformation of the fields S a ( ) and S a ( ) at = 0 is
therefore
S a ( ) = i
M ab S b ( ),
S a ( ) = i
M ab S b ( ).
(4.12)
(4.13)
Because of the presence of i, this transformation is not unitary. However, the appearance
of i here inherits the i in (3.3), (3.4), and therefore this should be regarded as an artifact for
considering the Euclidean D-brane. If it was possible to construct a boundary state of the physical
D-brane with manifest spacetime supersymmetry, then the scale transformation of fermionic
fields would also be an ordinary canonical transformation.
4.2. Boundary state
In this subsection, we examine the transformation property of the boundary state |B. First,
consider the transformation of non-zero mode oscillators in |B. Since each mode can be treated
separately, we consider, for n > 0, the transformation of

1 ij i j
ab a b
|bn = exp M n n iM Sn Sn |0n .
(4.14)
n
It is important to notice that |0n is also transformed by the scale transformation, since it mixes
the creation and annihilation operators as (4.6), (4.7). That is,
acts on |0n as a Bogoliubov
transformation. Let |0n be the transformed vacuum. The explicit form is

ij i j

ab a b
|0n = exp M n n i
M Sn Sn |0n .
n
217
(4.15)
The transformed state |bn is therefore
|bn = eAn n Hn e
An |0n ,
(4.16)
where
1
j
i
a b
An = M ij n
n iM ab Sn
Sn ,
n
i
i
a
a a
ni + n
ni + nSn
Sna + nSn
Hn = n
Sn .
(4.17)
(4.18)
Note that the constant terms, which could appear from the ordering of the oscillators, cancels
each other by the supersymmetry. Using the commutation relation
[Hn , An ] = 2nAn ,
(4.19)
it is easy to show that
|bn = eAn e n Hn e
An e
An |0n
= eAn |0n
= |bn .
(4.20)
Next, let us consider the zero modes. The transformation of zero modes is
S0a = i
M ab S0b ,
S a = i
M ab S b ,
0
(4.21)
(4.22)
which is consistent with (4.9). To examine the transformation of |B0 , let us construct the operator D which generates the transformation (4.21), (4.22). The explicit form of D is
D = M ab S0a S0b + c1,
(4.23)
where the transformation is defined as

=
etc. Here c is a constant which is not
determined simply by the algebra, since we consider only the subgroup SO(3, 1) SO(1, 1) of
the SO(4, 2) in our worldsheet description.
It is now straightforward to show that
S0a
|B0 = i
(4i + c)|B0 .
i
[D, S0a ],
(4.24)
Using the ambiguity of c, we can define the scale transformation so that |B0 is invariant. Since
the non-zero mode part of |B is also invariant, it has been shown that |B is scale invariant, for
the above definition of the transformation.
It is worth emphasizing again that the supersymmetry, as well as the symmetric property
of M ab , plays crucial roles for the invariance of the boundary state under the scale transformation.
If one considers a boundary state in, for example, bosonic string theory, then the variation of the
boundary state is proportional to itself with a divergent coefficient. By a regularization of the
divergence, one would obtain a non-vanishing term in the variation, even after subtracting a
divergent terms.
218
4.3. Variation of free energy

We would like to evaluate the variation of the free energy of a string in the D3-brane background under the scale transformation. The variation for the case of Fig. 1 can be evaluated
similarly, provided that we treat the transformation of the boundary C properly.
The free energy is defined as an infinite sum
F () =

Fn
n=0
n!
n ,
(4.25)
where Fn contains the contributions from worldsheets with n boundaries. The scale transformation
F () of the free energy is thus determined by
Fn . The quantity
Fn may have two
contributions, one of which comes from the variation of the action, the other of which comes
from the variation of the boundary. As shown in the previous section, each boundary is invariant
under the scale transformation. Therefore, we only need to consider the variation of the action.
Since the scale transformation is defined on the oscillators, it is convenient to consider in the
canonical formalism. The path-integral measure is

iSB +iSF ,
DXDP DSDSe
(4.26)
where

2
i 2
1
d 2 P i X i
+
P
X i
,
2
2

i a
i a
S S a + S a S a
S S a S a S a .
SF = d 2
2
2

SB =
Under the scale transformation, SB and SF transform as

SB = d 2
M ij P i P j + M ij X i X j ,
SF = d 2 M ab S a S b .
(4.27)
(4.28)
(4.29)
(4.30)
Performing the path-integral of P i , we can deduce the transformation of the fields in the Lagrangian formalism. It can be shown that the transformation is nothing but
X i =
M ij X j ,
S a = i
M ab S a ,
(4.32)
S a = i
M ab S a ,
(4.33)
(4.31)
and the variation of the action is therefore

d 2 M ij X i X i + M ab S a S a .
S =
(4.34)
The first term of the RHS can be regarded as a vertex operator of a massless graviton with zero
momentum. Note that this vertex operator does not contain the dilaton contribution, since M ij
is traceless, which is suitable for later discussion. The transformation of the supercharge (4.9)
implies that
S is invariant under Q , and therefore, the second term of the RHS of (4.34)
should be related to the vertex operator of an RR field.
219
Fig. 5. LSZ-like reduction. The cross represents the insertion of

S or |B0 .
It has been shown that the variation of the worldsheet action provides a linear combination of
vertex operators of massless fields, which are invariant under Q . Recall that the bosonic state
in the massless sector of the Hilbert space which is invariant under Q is proportional to |B0 .
Therefore, the variation
S of the action can be replaced with the insertion of the state |B0 .
Schematically, we denote

Fn =
Fn |B0 .
(4.35)
The tracelessness of M ij matches with the fact that the D3-brane is not a source of the dilaton.
To evaluate Fn (|B0 ), consider a worldsheet path-integral Fn+1 (z) with n + 1 boundaries,
one of whose boundary is placed at x i = zi for i = 5, . . . , 10. Note that the D3-branes we have
discussed so far are placed at x i = 0. In other words, we place another set of D3-branes which
are parallel to the original branes but their positions are different. It is possible to obtain Fn (|B0 )
from Fn+1 (z) through an LSZ-like procedure:

Fn |B0 = d 6 z z Fn+1 (z),
(4.36)
where z is the Laplacian on R6 . See Fig. 5 for an image of this procedure. The variation of the
total free energy is therefore

F () =
d 6 z z F (, z),
(4.37)
where
F (, z) =
n=0
n!
Fn+1 (z).
(4.38)
In the following, we will focus on a situation in which some operators placed around |x| = y
are inserted to the worldsheet. One example of such situations is to put a circular source of string,
as in Fig. 1. In this situation, the worldsheet is stretched in the region 0 |x| y. If we take |z|
to be larger than y, then the corresponding worldsheet has a thin tube connecting the boundary
and the body of the worldsheet. The tube represents the propagation of closed string states, and
the dominant contribution comes from the massless propagation. As a result, F (, z) behaves as
|z|4 for large |z|. For the range 0 |z| y, we assume that F (, z) varies slowly with |z|.
Let us calculate the integral

I = d 6 z z f (z),
(4.39)
where
f (z) =
f (0)
y4
|z|4
f (0)
(0 |z| y),
(|z| > y),
(4.40)
220
is a typical example of the function having the property assumed for F (, z). It is easy to check
that I = 4 vol(S 5 )y 4 f (0). Similarly, the RHS of (4.37) would be estimated as

d 6 z z F (, z) y 4 C(, y)F (, z = 0),
(4.41)
where C(, y) is assumed to be of order one. Noticing that Fn+1 (z = 0) = Fn+1 , we obtain
F ()
y 4 C(, y)
n=0
n!
Fn+1
=
y 4 C(, y) F ().
(4.42)
In other words, the free energy transforms as

F () F +
y 4 C(, y) ,
(4.43)
which indicates the existence of the scale invariance in the near-horizon region y 4 . Note
that the sum of the infinite number of worldsheets with boundaries is crucial for the existence of
this scale invariance.
The discussion so far is based on a power series expansion in terms of , and it is not obvious
whether the transformation property (4.42) also holds for large . To check the validity of (4.42)
for large , let us consider an effective sigma model description of this system. Since is large,
we can take y large enough so that the classical gravity approximation is valid (see Fig. 3), and
this fact verifies the use of the sigma model. The presence of the D3-branes is realized in the
sigma model by a non-trivial background field configuration, and the free energy of the model
would be a functional of the background fields. Let us focus only on the metric, for simplicity.
Then, due to (2.5), the free energy F [g] satisfies
F [g]
y4
F [g],
(4.44)
which is the same relation with (4.42) up to a factor of order one. This strongly suggests that the
free energy defined as a power series of can be continued analytically to the region of large
where the gravity description is valid, and therefore, the scale invariance in the near-horizon
region would exist for any .
So far, we have only considered the infinitesimal transformation. A remarkable property of
the scale transformation is that it brings the original free theory into another free theory which is
obtained via a field redefinition from the original one, as shown in (4.34). The boundary state |B
has the same form for both the original and the redefined field variables. Therefore, we can
perform the transformation repeatedly to obtain a finite transformation. The finite transformation
of the worldsheet fields is
X i ( ) eM
ii
X i ( ),
(4.45)
ab b
S ( ) cos S ( ) + i sin M S ( ),
(4.46)
S a ( ) i sin M ab S b ( ) + cos S a .
(4.47)
It is this fact that enables us to take the worldsheet of Fig. 2 to the one of Fig. 3 and vice versa.
221
Table 1
D3
D5
F1
Fig. 6. A worldsheet configuration for a supersymmetric Wilson loop. The string is stretched between D3-branes and
D5-brane. Open string states are emitted and absorbed by the string.
5. Supersymmetric Wilson loop

In this section, we show that the functional form of the supersymmetric Wilson loop is naturally derived from our worldsheet point of view.
A Wilson loop on the D3-branes was constructed, as mentioned in Section 2, by putting a
source of string in a target spacetime, and by stretching a string between them. For definiteness,
we choose a D5-brane as the source, which does not have common longitudinal spatial directions
with the D3-branes. The D-branes extend in the directions indicated in Table 1.
This brane configuration preserves 8 supercharges, which we denote as Q + Q where
and are restricted as
= 0 1 2 3 ,
(5.1)
= 0 5 6 7 8 9 .
(5.2)
Recalling that both and are spinors with, say, positive chirality, of SO(9, 1), these supercharges are preserved even after a string is stretched between the D3-branes and the D5-brane,
as depicted in Fig. 6. Therefore, the Wilson loop constructed here should preserve half the supersymmetry of N = 4 SYM. Its functional form is

1
I I .
W (C) = Tr exp i ds x A + |x|
(5.3)
N
C
This construction of the Wilson loop was also discussed in [8,12].

We would like to derive (5.3). The coupling of the gauge fields and the scalars on the D3branes to the worldsheet is described by the corresponding vertex operators
(p) X eipX ,
I (p) X I eipX ,
(5.4)
222
respectively. Recall that the mode expansion of the open string stretched between the D3-branes
and the D5-brane is
X = x + 2 p + ,
(5.5)
aI
XI = + ,
(5.6)
where a I is a vector indicating the separation of the D3-branes and the D5-brane, and the dots
indicate the oscillation parts of the expansion. For our choice of C, only the p 0 component is
non-zero. The Virasoro constraint implies
1 I
a .
This relation leads

2 p 0 =
(5.7)

(p) X + I (p) X I = 2 p 0 0 (p) + p 0 I i (p) + ,
(5.8)
where I is a unit vector parallel to a I . Therefore, the coupling induced by (5.3), including the
relative coefficient, is correctly reproduced from the worldsheet picture.
6. Discussion
We have shown that there is a scale invariance in the worldsheet theory of type IIB string with
D3-branes as far as closed string loops are suppressed, if we only consider in the near-horizon
region. This scale invariance actually implies the AdS/CFT correspondence for the Wilson loops
in N = 4 SYM on the D3-branes and the minimal surfaces in the corresponding AdS5 S 5
spacetime. In the gauge theory point of view, we take the large N limit, and we choose a large
t Hooft coupling when we compare it with the results from classical gravity. Therefore, we
have shown a part of the weakest version of the AdS/CFT correspondence. Our point of view
also provides a natural understanding of the functional form of the supersymmetric Wilson loop
operator in terms of the worldsheet stretched between the D3-branes and a D5-brane.
It is desired to examine whether it is possible to argue a more stronger version of AdS/CFT
correspondence in the worldsheet point of view. Since this is a perturbative description, the string
coupling must be taken to be small. As discussed at the end of Section 2, this is also crucial
for considering the near-horizon region appropriately. The number N of the D3-branes is not
necessarily large in the worldsheet description. However, if N is finite, then must be small,
and the near-horizon region is also small, as shown in Section 4. If the width of the near-horizon
region is of order of the string scale, then worldsheet may fluctuate and reach the outside of
the near-horizon region, and the scale invariance will not be available. It seems natural to think
that the ordinary claims of AdS/CFT might be modified in this situation. It is possible to take
a large N limit, and obtain a finite which is bigger than the string scale, but not big enough
for the classical gravity description to be valid. In this case, it would be necessary to include
-corrections, and our point of view may provide a guideline for how to include the corrections.
The strongest version, that is, both N and are finite, is beyond the reach of our point of view,
and a non-perturbative formulation of string theory will be necessary to discuss this issue. Of
course, there might be a possibility of occurring miraculous cancellations among worldsheets
due to supersymmetry, and as a result, the scale invariance might be realized more generally. The
range of validity of AdS/CFT correspondence in a stronger version was also discussed in [13].
223
Fig. 7. GKPW relation.
It might be possible to apply the line of our arguments to less symmetric cases. It seems that
one of the significant differences from the N = 4 version will be the transformation property of
boundary states. To illustrate the difference, let us consider a boundary state in bosonic string
theory

1
j
ij i
|B = exp
(6.1)
M n n |0,
n
n=1
where i and j run from 1 to 24, and M ij is suitably defined. The transformation of the oscillators
will be the same as (4.6),
(4.7). A naive calculation shows that the transformed state |B is
proportional to exp[24

n=1 1], and therefore some regularization is necessary. We employ
the regularization

|B e
H |B,
(6.2)
where H is the Hamiltonian, and take

0 limit at the end of calculations. The result is
12
|B 12(H + 1)|B.
(6.3)

The finite part of the RHS has an interesting form, since this would be rewritten as a derivative of
the boundary state with respect to a moduli parameter. This would imply that the boundary of the
moduli space of Riemann surfaces with boundaries would be relevant for evaluation of
F ().
This might be important when one considers a case with a generic D-branes.
In our argument, the only possible subtle point would be the estimate (4.41). Although we
could not completely verify the validity of (4.41), we think it is already a great achievement
that we could reduce AdS/CFT correspondence to a concrete problem of evaluating the LSH
of (4.41). We expect that our approach presented in this paper will provide a way of proving
AdS/CFT correspondence, since it is now reduced to a problem in a free theory.
It would be plausible if our argument can be extended to apply to the examination of the
GKPW relation [9,10]. Naively, it should be possible, since the Wilson loop is a source of generic
string states, and therefore, the consideration of multiple Wilson loops would implies the GKPW
relation. An intuitive image of how to show the GKPW relation is depicted in Fig. 7. To describe
an insertion of a BMN operator [11] on the D3-branes, for example, it would be suitable to
choose a Euclidean D3-brane which is perpendicular to the original D3-branes as a source of
string. See Fig. 8. We hope to report on these issues elsewhere.
|B =
224
Fig. 8. Insertion of a BMN operator.
Acknowledgements
We would like to thank T. Azeyanagi, Y. Matsuo for valuable discussions. This work is supported by the Grant-in-Aid for the 21st Century COE Center for Diversity and Universality
in Physics from the Ministry of Education, Culture, Sports, Science and Technology (MEXT)
of Japan. The research of T.S. is supported in part by JSPS Research Fellowships for Young
Scientists.
References
[1] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[2] S. Rey, J. Yee, Macroscopic strings as heavy quarks: Large-N gauge theory and anti-de Sitter supergravity, Eur.
Phys. J. C 22 (2001) 379, hep-th/9803001.
[3] J.M. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
[4] M.B. Green, M. Gutperle, Light-cone supersymmetry and D-branes, Nucl. Phys. B 476 (1996) 484, hep-th/9604091.
[5] J.K. Erickson, G.W. Semenoff, K. Zarembo, Wilson loops in N = 4 supersymmetric YangMills theory, Nucl. Phys.
B 582 (2000) 155, hep-th/0003055.
[6] N. Drukker, D.J. Gross, An exact prediction of N = 4 SUSYM theory for string theory, J. Math. Phys. 42 (2001)
2896, hep-th/0010274.
[7] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, vols. 1, 2, Cambridge Univ. Press, 1987.
[8] S. Yamaguchi, Wilson loops of anti-symmetric representation and D5-branes, JHEP 0605 (2006) 037, hepth/0603208.
[9] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105, hep-th/9802109.
[10] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[11] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-YangMills,
JHEP 0204 (2002) 013, hep-th/0202021.
[12] J. Gomis, F. Passerini, Holographic Wilson loops, JHEP 0608 (2006) 074, hep-th/0604007.
[13] I.Y. Park, Fundamental vs. solitonic description of D3 branes, Phys. Lett. B 468 (1999) 213, hep-th/9907142.
Non-Abelian magnetized blackholes and unstable

attractors
A.E. Mosaffa a , S. Randjbar-Daemi b , M.M. Sheikh-Jabbari a,
a Institute for Studies in Theoretical Physics and Mathematics (IPM), PO Box 19395-5531, Tehran, Iran
b The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11 34014, Trieste, Italy
Received 30 April 2007; received in revised form 25 July 2007; accepted 10 August 2007
Abstract
Fluctuations of non-Abelian gauge fields in a background magnetic charge contain tachyonic modes
which as we will show cause an instability of the background. We extend this result to the cases where the
background charge (flux) is coupled to four-dimensional Einstein gravity and show that the corresponding spherically symmetric geometries, which in the absence of a cosmological constant are of the form
of (colored) ReissnerNordstrm blackholes or the AdS2 S 2 , are also unstable unless the flux assumes
its smallest allowed value, in which case the configuration is stable. We discuss the relevance of these
instabilities to several places in string theory including various string compactifications and the attractor
mechanism. Our results for the latter imply that the attractor mechanism shown to work for the extremal
Abelian charged blackholes, cannot be applied in a straightforward way to the extremal non-Abelian colored
blackholes, with the exception of the minimally charged stable ones.
1. Introduction and motivation

Stability analysis of systems of charged particles coupled to Einstein gravity in four dimensions is an old and well-studied subject. Static, spherically symmetric solution with electric
and/or magnetic charge, the ReissnerNordstrm blackhole, has been shown to be stable under classical stability test. Charged or neutral scalars can also be added to the EinsteinMaxwell
E-mail addresses: mosaffa@theory.ipm.ac.ir (A.E. Mosaffa), seif@ictp.trieste.it (S. Randjbar-Daemi),

jabbari@theory.ipm.ac.ir (M.M. Sheikh-Jabbari).
doi:10.1016/j.nuclphysb.2007.08.003
226
A.E. Mosaffa et al. / Nuclear Physics B 789 (2008) 225244
theory. A special case of the latter is the EinsteinMaxwellDilaton (EMD) theory. String theory
toroidal compactifications at generic points in their moduli space indeed lead to an EMD theory with several U (1)s as well as dilatonic fields (e.g., see [1]). All these generalized charged
blackholes are stable.
As the next generalization, one may couple non-Abelian gauge theories to Einstein gravity.
These theories are hence called EinsteinYangMills (EYM) theories [2]. Such theories have
also been studied in some detail and it has been shown that they have an infinite discrete family
of globally regular solutions as well as blackhole solutions for any given value of the horizon
radius [35]. Despite having time independent globally regular or (colored) blackholes solutions,
it has been argued that these solutions are generically unstable. (For a more detailed review
see [6].)
When Dilaton or Higgs scalars are added to the system we obtain, respectively, EYMD or
EYMH theories. These are the kind of theories which can arise from string theory compactifications, though not in generic points of the moduli space of toroidal compactifications. They also
arise from compactification on group manifolds or generic CalabiYau manifolds. As we are interested in gravity theories arising from string compactifications, in this note we shall focus on
these theories and review some of the results regarding the stability of such magnetically charged
blackholes in Section 2.2. We extend the results on the instability of the colored blackholes to a
wider class in which the unstable (tachyonic) mode is not an s-wave.
In the rest of this note, we will first review the results of [7,8], stating the existence of the
tachyonic modes in the fluctuations of the YangMills fields in a monopole background. In
Section 2.1, we complete the existing literature by showing that indeed the existence of these
tachyonic modes leads to the instability of the background. In Section 2.2, we review arguments
in the GR literature in which it has been shown that a generic class of solutions to the EYM
system, namely the non-Abelian regular or blackhole solutions, are unstable. In Section 3, we
focus on the instabilities of the magnetic monopole backgrounds, the embedded Abelian configurations in the GR terminology, in the presence of gravitational field. In this section we show
that if a YangMills configuration has destabilizing modes among its fluctuations already in the
flat space, then coupling to gravity will not change the situation as regards to stability. To this
end first we show that the Dirac monopole in an EYM(D) theory generically has fluctuations
which are localized in space but grow exponentially in time. With a given monopole configuration or the corresponding magnetic flux, the space time geometry is either of the form of (colored)
ReissnerNordstrm (RN) blackhole or AdS2 S 2 . We show that both of the two cases are suffering from a similar kind of instability present in the non-gravitational case. The solutions with
minimal monopole charge are, however, an exception; they are stable. This seems to have been
overlooked in previous analysis.
For the blackhole solutions, our results extend the known instability arguments of the GR
literature to a larger class, and in particular to the cases different from s-wave perturbations and
monopole charges different from minimal value for the tensorial perturbations. The AdS2 S 2
is of relevance to the attractor mechanism which is at work for extremal blackhole solutions in
the EinsteinMaxwellDilaton (EMD) theories (e.g., see [9,10]). As we discuss in Section 4,
the instability of the (extremal) colored blackholes and its near horizon geometry, the AdS2
S 2 background, implies that, if applicable at all, the attractor mechanism in its present form
does not work for the colored non-Abelian blackholes. In Section 5, we discuss that the EYMD
or EYMH theories are very relevant to string theory and therefore the results of the colored
blackhole instabilities apply to string theory compactifications. We end with concluding remarks
and some interesting open problems.
227
2. Review of instabilities in background non-Abelian magnetic flux

This section is mainly a review of known results. In Sections 2.1 and 2.2, respectively, we consider the non-gravitational and gravitational cases and point out that in both cases the fluctuations
of non-Abelian gauge fields about a monopole background contain unstable modes.
2.1. Linearized analysis of the gauge fields fluctuations
Fluctuations of a non-Abelian YangMills field around a magnetic monopole background
on R 3+1 which has a non-vanishing flux through S 2 , has tachyonic modes which can create
instabilities [7]. To study such fluctuations it is sufficient to study the linearized YangMills
equations DM F MN = 0, where F MN is the gauge field strength, around the solution of interest.
To proceed we use the notation M = , m, where ranges over x 0 = t and x 1 = r and m ranges
over and . We also take the four-dimensional metric of the form
ds 2 = g dx dx + r 2 gmn dy m dy n ,
(2.1)
where gmn denotes the standard metric on an S 2 of unit radius. Next we write
WM = A M + VM ,
(2.2)
where WM is the gauge field potential and A M is the background potential corresponding to a
Dirac monopole located at the origin (and hence has spherical symmetry)
1
A = ni Ti (cos 1) d,
(2.3)
2
where Ti is a basis of matrices in the Cartan sub-algebra of the Lie algebra of G and, in an
appropriate normalization, ni which correspond to the charge(s) of the monopole are integers.
YangMills equations, up to linear order in Vn , give,

1 2 n
D V Dm , D n V m i Vm , F mn = 0.
(2.4)
r2 m
In the above the covariant derivatives and m are respectively constructed using the metrics
g and gmn and the m, n and , indices are raised and lowered by metrics gmn and g . Dm
is defined by
2 V n +
Dm Vn = m Vn i[A m , Vn ],
(2.5)
and
ni
F = T i sin d d.
2
The above equations of motion are written in the Lorentz gauge
(2.6)
1
Dm V m = 0.
(2.7)
r2
Our interest is primarily in the equation for V n . We can express the commutators in terms of the
curvature tensor. The equation simplifies to
V +
2V n +

1 2
Dm 1 V n 2i Vm , F mn = 0.
2
r
(2.8)
228
Expanding V n (t, r, , ) in harmonics on S 2 yields an infinite number of fields in the (1 + 1)dimensional space spanned by t and r. The general formalism has been given in [11]. To perform
such an expansion it is necessary to use complex basis in the tangent space of S 2 and denote the
component of Vn with respect to such basis by V+ and V . The equations then separate and we
obtain

1 2
1 V+ 2i[V+ , F+ ] = 0.
2 V+ + 2 Dm
(2.9)
r
Writing
V+ = V+i Ti + V+a Ta ,
(2.10)
where Ti , Ta are, respectively, the generators in and outside the Cartan subalgebra satisfying
[Ti , Ta ] = ia Ta ,
(2.11)
the equation for V+a becomes

1 2
Dm 1 V+a + ni ia V+a = 0.
2
r
This equation can be written in the more suggestive form of
2 V+a +
2 V+a
1 2
M V+a = 0,
r2
where the mass operator M 2 is defined by

2

1 + ni ia .
M 2 = Dm
(2.12)
(2.13)
(2.14)
The spectrum of this operator is known. We know that for the non-Abelian gauge group G
being any classical group and their products M 2 has a single negative eigenvalue (generically
corresponding to degenerate eigenstates) and an infinite number of positive eigenvalues [7,8].
The multiplicity of the mode with negative M 2 , tachyonic modes,1 is given by the dimension
of the representation of the overall unbroken group to which they belong. For example, assume
that the gauge group is SU(2). The background gauge field is that of a monopole of charge n
which we take it to be a positive integer greater than one in the direction of T3 . The reason we
exclude n = 1 is that there are no negative M 2 modes for this value of n. In this case all the
tangential fluctuations of the gauge field are spinorial and the Abelian embedding is stable under
small fluctuations.
In this example, if there are no other symmetry breaking elements (like charged scalar
fields or the Higgs fields) the unbroken part of the symmetry group in the t , r space will be
U (1) SU (2)KK , where, U (1) is the unbroken subgroup of the gauge group SU(2) and the factor SU(2)KK , is due to the spherical symmetry of the background solution. The tachyons in that
case are charged under the U (1) and belong to the spin j = n/2 1 representation of SU(2)KK .
This in particular means that for even n the tachyons are in tensorial representations of the rotation SU(2)KK . Note that for odd n the tachyons are spinorial objects with respect to SU(2)KK . The
1 Note that if the 2-dimensional metric is flat, i.e., g
= , and the 4-dimensional spacetime is a product of
R 2 S 2 then M 2 will be precisely the mass2 operator in R 2 and the appellation of tachyonic mode will be exact. By
slight misuse of language we shall continue to call M 2 the mass operator even for a non-flat g and for non-factorized
geometries.
229
s-wave tachyon corresponds to n = 2. Moreover there are of 2j + 1 = n 1 complex tachyons

2 = j [7]).2
(for n > 1) with M 2 = n/2 (for the tachyonic modes Dm
2
One should, however, note that having negative M is not in itself sufficient to conclude the
instability of the field configuration, one needs also to analyze the propagation of these negative
M 2 modes in the t , r space.
If we assume g = , the fluctuation equation (2.13) reduces to the simple wave equation
1 2
M V+ = 0.
r2
The most general solution to this equation turns out to be

V+ = eiEt eiEr r C1 F (, 2; 2iEr)
t2 V+ + r2 V+

+ C2 (2iEr)12 F (1 , 2 2; 2iEr) ,
(2.15)
(2.16)
where F (a, c; x) is a confluent hypergeometric function, C1 and C2 are the two integration constants and is a solution of
( 1/2)2 = M 2 + 1/4.
(2.17)
For M 2 < 1/4, which is the case of interest for us, the right-hand side of this equation can be
negative yielding a complex solution for . Assuming this situation

1
1 1
= M 2 + i ( is real)
(2.18)
2
4 2
and taking the plus sign in the above (the case with a minus sign is quite similar) we can have
solutions with an exponential growth in time. To show this, we note that with of the form given
in (2.18) confluent hypergeometric function can be expressed in terms of Bessel functions

V+ = e t r 1/2 C1 4i (1 + i)Ii ( r) + C2 4i (1 i)Ii ( r) ,
where we take = iE to be a positive real number. Demanding that the solution vanishes at large
values of r, to ensure the normalizability of the mode, we find
(1 i) 2i
C1
=
4
C2
(1 + i)
(2.19)
(otherwise it grows as e r ). The condition (2.19) brings the solution into the following form
V+ = e t r 1/2 Ki ( r)
(2.20)
with K being the modified Bessel function. At small values of r, the function behaves as

(1 + i)
t 1/2
.
V+ e r sin ln (2 r) + 2 ln 2 + /2 , = arg
(1 i)
The solution (2.20) is hence clearly normalizable and localized in space but delocalized (with
exponential growth) in time. This establishes the instability. We would like to emphasize that as
is seen from the above analysis, M 2 < 1/4, and not just negative M 2 , cause the instability.
2 A similar instability also exists if instead of monopoles we have vortices, the NielsenOlesen instability [12]. This
instability in the context of string theory have also been studied in [13].
230
If there are other charged fields in the model they may contribute to the tachyonic mass and
lift it to non-negative values. One such possibility is adding the Higgs field in the adjoint (the
GeorgiGlashow model) or the fundamental representations. For the former, there are singular
or regular (t HooftPolyakov) monopole solutions.
Let us consider the GeorgiGlashow model, with the group SU(2) and a Higgs potential
whose value vanishes in its minimum given by |vac |2 = v 2 (e.g., see Section 1.4 of [14]):
3
2
1

1 i
i MN
i M i
+ DM D 2 v 2
FMN F
L=
4
2
4
(2.21)
i=1
where
j
i
= [M WNi ] gYM ij k WM WNk ,
FMN
j
DM i = M i gYM ij k WM k .
(2.22)
Our arguments in an obvious way generalize to a generic non-Abelian gauge group G. The
singular monopole solution is given by (see Eq. (23) of [14])
vac vac = v 2 ,
M = 1 vac M vac + 1 vac A0 ,
W
M
v
gv 2
(2.23)
where the arrows denote the (adjoint) gauge indices and A0M is an arbitrary vector, and singlet
M vac = 0. The
of the gauge group. With the above it is easy to check that M vac + ig W
magnetic monopole U (1) gauge field strength is then given by FMN = v1 vac FMN . This
monopole solution is obviously singular. We choose the three components of to be given by
v(sin cos m, sin sin m, cos ), m Z.
The stability analysis in this case has been carried out in [15] for the s-wave perturbations
(which corresponds to m = 1 in this case). In fact the gauge transformation which brings to
the direction of z-axis will map W to our A of (2.3) with m = 2n.3 Repeating the analysis of
the spectrum of gauge field fluctuations about the above when Higgs is turned on, will therefore
produce the same result given by Eq. (2.13) but the M 2 is now replaced with
2
M 2 = M02 + r 2 gYM
v2,
(2.24)
where M02 is the expression given in (2.14) and the second term is due to the Higgs giving mass to
the gauge fields. Depending on whether our Higgs field is in the adjoint (as in the above) or in the
fundamental (like the Standard Model Higgs) the value of the monopole charge n is respectively
even integers or an integer. It can be checked that for small enough r, r rm , with
rm (gYM v)1
(2.25)
the last term becomes subdominant and we again find the tachyonic instability.
Finally we consider the regular t HooftPolyakov solution. The (approximate) form of solution may be found in Eq. (29) of [14]. This solution at large r reduces to (2.23). At r rm ,
3 It is known that with the adjoint Higgs one will only obtain Dirac monopoles of even charge. In particular the lowest
possible charge in this setting corresponds to n = 2 in the notations of (2.3). This is the reason that our s-wave is obtained
for n = 2.
231
however, the solution behaves differently and in particular at r = 0 it is regular. In this case
the monopole is an object of effective radius of rm , rather than a point-like object. The stability
analysis for this case, due to the r-dependence of the background Higgs and the gauge fields becomes very messy, however, one expects this solution to be stable. This expectation is supported
by topological reasons and, in our setting, by the observation that due to the monopole profile, the
M 2 has a non-trivial r dependence and unlike the previous case, does not change sign around rm .
Similar arguments could be made for the fundamental Higgs case and when we have dyons.
2.2. The gravitational setting, instability in EYM(D) theories
In this section we give a brief review of the non-Abelian gravitating solutions of EYMD theory
in four dimensions (for a more detailed review see [6] and references therein). These solutions
can be either point-like, globally regular, asymptotically flat (solitonic) or of the blackhole type.
There are a number of no-go theorems assuring that stationary solitonic solutions do not exist in
pure gravity or in YM theory. In gravity, this is referred to as the Lichneroviczs theorem which
follows from the fact that gravitational systems are purely attractive and hence there is no way to
make a balance of forces to keep the gravitational soliton stationary. On the other hand, the YM
system is purely repulsive and therefore it cannot support what is called a classical glueball.
When combined together, gravity and YM can in principle make a balance between forces to
support stationary neutral solitonic4 [2] or neutral and charged blackhole solutions [4]. Similar
class of solutions have also been constructed for systems including a dilatonic scalar field, i.e.,
EYMD [5].5
In this section we will take the gauge group to be SU(2) and consider four-dimensional,
spherically symmetric, asymptotically flat static solutions only. To summarize the results already
discussed in the literature and introduce the terminology used, it is convenient to start with the
most general ansatz for spherically symmetric solutions

1 2
dr R 2 d 2 + sin2 d 2 ,
N
n
n
W = aT3 + Im(wT+ ) d Re(wT+ ) sin d + T3 cos d,
(2.26)
2
2
where Ti (i = 1, . . . , 3) are the SU(2) generators and T+ = T1 + iT2 . Furthermore, a is a real
one form a = a0 dt + ar dr and the complex scalar w as well as , N, and R only depend on
(t, r). The integer n is the monopole charge.
A study of the equations of motion for the YM field indicates that for n
= 2 one should
necessarily have w = 0. Solutions of this type are called embedded Abelian (eA), as the potential
in this case can be written in terms of the U (1) potential A = Tr(W T3 ) = a + n2 cos d. The
stationary (eA) solutions are hence, always charged and are blackholes, the so-called colored
blackholes.
On the other hand, the non-Abelian (nA) solutions which can only be achieved when n = 2,
may be solitonic or blackholes, but always neutral. The stability of these (nA) solutions against
spherically symmetric perturbations has been extensively studied in the literature with the overall result that these solutions are all unstable. There are two classes of the unstable, tachyonic,
ds 2 = 2 N (dt + dr)2
4 There is a no-go theorem stating that there exists no charged soliton in EYM [16].
5 Systems including scalar fields in the non-trivial representation (fundamental or adjoint) of the gauge group, the
Higgs field, have also been extensively studied under the name of EYMH, e.g., see [15,17,18].
232
modes; those in which the gravity fluctuations are turned off, the topological or odd parity modes,
and those which also involve metric perturbations, the gravitational or even parity modes.
The above system can be generalized to the one which is more relevant to string theory, by
adding a dilaton field to the EYM to end up with the EYMD system. The dilaton, , couples to
the YM action in the form of e2 LYM where 0 is a parameter. For any given , the (nA)
solutions of the EYMD system as well as their stability behavior are the same as those of EYM,
i.e., there are both odd and even parity tachyonic modes [5,19]. The extreme limits of 0
and respectively correspond to the EYM and YMD (in flat space). The latter which is
essentially the case we analyzed in Section 2.1 has the same features of EYM, i.e., the dilaton,
providing a purely attractive system, can replace gravity leading to the same sort of solutions.
The EYMD will thus interpolate between these extreme systems and has the same features of the
two.
Another system of interest is an EYM with a cosmological constant; EYM- [20]. The solutions of this system are asymptotically (A)dS for (negative) positive values of . For > 0,
the moduli space of (nA) solutions is essentially that of the EYM system and all the solutions
are still unstable, with the same tachyonic modes. For < 0, however, the situation is different.
The main difference is that, the presence of negative allows for a new continuous family of
non-Abelian charged solitonic or blackhole solutions. The stability analysis for this case shows
that [21,22], although the solutions are still generically unstable, we now have the possibility of
having stable solutions which of course do not have any counterpart in the = 0 theory.
3. Instability of EYM(D) solutions with magnetic flux
In the previous section we reviewed the instability of the so-called non-Abelian solitonic or
blackhole solutions. In this section we apply the analysis of Section 2.1 to all possible spherically
symmetric solutions to the EYM action

1
4
MN
S = d x g
(3.1)
R + Tr FMN F
8GN
when the magnetic flux Fmn (2.6) is turned on. With the gauge fields of the form of (2.6) and noting that F only takes values in the Cartan subalgebra of the gauge group G, finding solutions of
the field equations is essentially identical to the one in EinsteinMaxwell theory. These solutions,
if asymptotically flat, are of the form of (magnetized) ReissnerNordstrm (RN) blackholes, the
so-called embedded Abelian colored blackholes, of arbitrary magnetic charges ni . There is only
one other class of solutions which are not asymptotically flat. These are non-singular and of the
form of AdS2 S 2 . The latter may be obtained as the near horizon geometry of the extremal RN
solutions.
We show that turning on gravity does not remove the instabilities we discussed in Section 2.1
and the same modes of the gauge field which were tachyonic in the non-gravitational case6
remain tachyonic. Hence, we are led to the conclusion that
the presence of gravity is irrelevant to the instability results of the non-gravitational case.
Inclusion of dilatonic scalar fields does not alter this result. Addition of Higgs scalars, being
6 In the RN solutions generically there are four parameters (or quantum numbers) appearing, the Newton constant G ,
N
the YangMills coupling gYM , and monopole charges ni and the ADM mass of the blackhole. By non-gravitational limit
we mean GN 0 keeping the other parameters fixed.
233
a charged field under YM, can however push some or all of the tachyonic modes inside the
horizon and thus may stabilize the system.
In Section 3.1 we analyze the (in)stability of embedded Abelian, colored blackhole case. In
comparison to the extensive literature on the non-Abelian solutions, the literature on the embedded Abelian case is slim and only the case with s-wave GerogiGlashow monopole has been
considered in detail, showing that such perturbations lead to instabilities [15].7,8
3.1. Instability of generic 4d colored blackholes
The (in)stability analysis of a colored, magnetized RN blackhole amounts to studying (2.13)
with the proper choice for the metric g . In order that we take the metric of the form
ds 2 = f dt 2 + f 1 dr 2 + b2 d22 ,
(3.2)
where f and b are functions of r. Eq. (2.13) written with this metric yields
f 1 t2 V+ + r (f r V+ )
1 2
M V+ = 0,
b2
(3.3)
where M 2 has the same expression as in (2.14). For concreteness, we take the gauge group to be
SU(2) for which the spectrum of M 2 has a simple form. To analyze the above equation, define
the new coordinate
d = f 1 dr.
(3.4)
In this new coordinate system the metric becomes
ds 2
= f (dt 2
+ d 2 ).
Replacing
V+ = eiEt (),
Eq. (3.3) then takes the form of a Schrdinger equation
2 + U () = E 2 ,
with the energy
E2
(3.5)
and the potential U ()
f 2
M .
(3.6)
b2
The question of instability, like the flat space case, therefore reduces to the existence of normalizable, negative E 2 states (the bound states). The normalizability condition should now be
imposed taking the non-flat background metric into account. That is,

d ||2 < .
(3.7)
U () =
7 Ref. [15] also considers some tensorial non-spherically symmetric modes. As discussed in Section 2.1, however, for a
generic monopole of charge n, e.g., in the SU(2) gauge theory, the spin of the tachyonic mode is n/2 1. Our discussion
applies to modes belonging to any representation of the rotation SU(2), tensorial as well as spinorial.
8 The case of electrically charged solutions to EYM and their stability has been addressed in [23]. This system has
also unstable modes, like ours, are in the directions transverse to the Cartan in the internal gauge group and which
unlike our case, are in the component of the gauge field along the radial direction of the spacetime. These modes are
specified by their angular momentum l on the sphere. For given background charge, the low angular momentum modes
are unstable [23].
234
Fig. 1. The potential given in Eq. (3.6) for the extremal ReissnerNordstrm blackhole, with M 2 = n/2. This potential
has negative energy bound states.
For the asymptotically flat case, like ours, at large r (where the normalizability concerns may
arise) the measure reduces to that of standard two-dimensional flat space dr. (One should also
note that functions f and b should be written in terms of .) In the cases of interest for us we
are dealing with blackholes and hence the function f (r) vanishes at the horizon and generically
we have inner and outer horizons, respectively r and rh . The Schrdinger equation should be
analyzed for r rh . The wave-function should vanish for r < rh , as the region inside the
horizon is cut-off from the rest of the space.
As the first example let us consider the extremal ReissnerNordstrm (RN) case, where

Q 2
f = 1
(3.8)
,
b = r.
r
Q and the monopole charge n are related as
n
n
4GN
M 1 .
Q=
2gYM
2gYM Pl
(3.9)
In this case the explicit relation between and r is

= r Q + 2Q ln
Q2
r Q
,
Q
r Q
(3.10)
where, for convenience, we have chosen the integration constant such that r = 2Q corresponds
to = 0. Note that ranges from to + as r ranges from the horizon rh = Q to +. This
coordinate does not cover the region inside the horizon. The (, t) coordinate system is usually
called the ReggeWheeler coordinates.
It is easy to see that = are inflection points of the potential and that U vanishes at
these points. Furthermore, U () has one more extremum at = 0 which is a minimum for
M2
negative M 2 and a maximum for positive M 2 and its value at this extremum is 16Q
2 . U () is
smooth everywhere and U () = U (). The case of interest for us is of course when M 2 is
negative which is depicted in Fig. 1.
Qualitatively our quantum mechanical problem is very similar to the motion of a particle in
2
a potential of the form cosMh2 (the latter has been discussed in [24]). This potential has a finite
2
M
M
2
2
number of normalizable discrete modes (bound states) for 16Q
2 E 0. For E 16Q2 of
course there will be no solutions. Existence of negative energy bound states, as discussed earlier,
means that the unstable modes, i.e., the modes which grow exponentially in time, which were
235
Fig. 2. The potential for generic RN blackhole with the Higgs. As we increase the Higgs expectation value v the minimum
of the potential is pushed to the left and eventually can be removed. The figure on the left shows the case with a large
Higgs and the right one shows the case with small v.
also present in the flat background survive even when the back reaction of the magnetic monopole
on geometry is taken into account.
One may repeat the above analysis for more general non-extremal cases where
1
(r r )(r rh ).
r2
The qualitative behavior of the potential is essentially the same as the extremal case. That is, for
negative M 2 the potential U () has a minimum outside the horizon, and it is always negative.
Furthermore, in the , U () asymptotes to zero.
Let us now add the non-singlet scalar field, the Higgs field, which can be in the fundamental
or in the adjoint representation of the gauge group. As discussed in Section 2.1, this is done by
2 v 2 b2 (cf. Eq. (2.24)). U () becomes,
replacing the M 2 by M 2 + gYM

(r r )(r rh ) 2 2 M 2
gYM v + 2 ,
U () =
(3.11)
r2
r
f (r) =
where v is the Higgs expectation value and and r are related as in (3.10). The potential U (),
unlike the previous case does not necessarily have bound states. As the expectation value of the
Higgs field increases, the minimum of U () is pushed behind the horizon. This can eventually
result in a potential without any bound states and hence large Higgs vev can remove the instability.9 Intuitively and as a rough measure, when we add Higgs there are two length scales in the
problem, one is the horizon size rh and the other is the monopole size rm (2.25). When rh > rm
the monopole is completely sitting behind the horizon and hence the instability analysis is similarly to a point charge. In this case, although the potential has a minimum, its minimum is
sitting behind the horizon and hence do not form a bound state. Therefore, we do not see the
instability. When rm > rh we have a smoothed out, extended monopole. As has been depicted in
Fig. 2 in this case the potential does have the possibility of bound state, and hence the instability.
The same qualitative features remain when we consider non-extremal case with the Higgs. The
above potential generalizes the results of [15] to the case with arbitrary monopole charge.
It turns out that bound states and hence unstable modes appear if M 2 < 0 and if
2
c M 2 MPl
n2
v
gYM
Mbh <
+
,
(3.12)
2
v
c M 2 4
9 This observation had already been made in [15] for the s-wave instabilities. It is also worth noting that the n = 1 case,
for which M 2 > 0, is stable. The minimal charge case of Ref. [15] corresponds to n = 2 in our notation (cf. footnote 3)
which is the lowest n the instability occurs.
236
+rh
where MPl is the Planck mass and Mbh = r2G
is the ADM mass of the blackhole. c is a positive
number smaller than one and arises from the details of quantum mechanics analysis of having a
bound state. For the extremal case, Mbh G = Q, c approaches one.
3.2. Instability in AdS2 S 2 background

In this subsection we consider the other class of solutions to the EYM theory in which the
field strength is given by (2.6), the AdS2 S 2 . This solution can be obtained as the near horizon
geometry of extremal colored blackholes and hence this case is of particular interest because
of its possible relevance to the attractor mechanism [25,26]. In this section we show that the
instability we have discussed in the previous sections is also seen in the near horizon geometry.
It is straightforward to check that the AdS2 S 2 :

ds 2 = R12 f (r) dt 2 + f 1 dr 2 + R22 d22 ,
(3.13)
where R1 and R2 are, respectively, radii of the AdS2 and S 2 parts, and f (r) is
f (r) = r 2
for the Poincar patch,
f (r) = 1 + r
in the global AdS coordinates,
(3.14)
when R1 = R2 = Q is a solution to our EYM theory. This solution can be obtained in the near
horizon limit of the four-dimensional extremal ReissnerNordstrm blackhole of previous section. Although for our case, the EYM theory and when four-dimensional cosmological constant
is zero, always R1 = R2 , in our analysis we keep the AdS and the sphere radii arbitrary.
Our instability analysis goes through the same as before. We end up with looking for the
possible bound states of the Schrdinger equation (3.5), in which the potential is given by (3.6)
with b replaced by R2 . That is,10
U () =
2
R
R12 M 2 12 ,
< < + in the Poincar patch,
2 < < + 2
2
R12
R22
M 2 cos12 ,
(3.15)
in the global AdS coordinates.
To explore the bound states and discuss the instability we need to impose the correct normalizability and boundary conditions. In the case where we have a causal boundary, like the AdS2
case, the suitable boundary condition for the wave function is that of a particle inside a box,
with the walls of the box at the position of the boundary, i.e., = 0 in the Poincar patch and
= /2 for the global AdS2 . The form of the potential for negative M 2 is depicted in Fig. 3.
The normalization conditions read

d ||2 <
for the Poincar patch,
10 Note that the AdS case is special in the sense that it has two disconnected one-dimensional boundaries [27]. In our
2
case this means that the range of coordinate can be extended to the full range of (, +) in the Poincar patch, and
to ( 2 , + 2 ) in the global AdS2 . The boundaries are then located at = 0 for the former and at = 2 for the latter.
Fig. 3. The potentials for the AdS2 case when M02 =
R12
R22
237
M 2 is negative. The left figure corresponds to the Poincar patch
(as well as to the flat space) and the one on the right corresponds to the global coordinate.
and
/2
d ||2 < in the global AdS coordinates.
/2
The above Schrdinger equation is in fact exactly the same as the wave equation for a massive
scalar field of mass squared M02 =
R12
M2
R22
in the AdS2 . Normalizable negative energy states exist
if M02 < 1/4. To see this, let us for example consider the Schrdinger equation for the Poincar
coordinates and focus on the equation around = 0. The solutions can be of the form ,
where satisfies (3.3) with M 2 replaced by M02 , i.e.,
( 1) = M02 .
(3.16)
As in the flat space case in order to have instability, should be a complex number.
In order to have a more complete stability analysis let us also discuss the AdS2 background
in the global coordinates and in presence of the above mentioned tachyonic mode. We use the
well-known result for the solution of the wave equation in this metric for a scalar with mass M0

1
t
2
= e sin F a, b; ; sin e t
(3.17)
2
with given in (3.16) and F being the hypergeometric function and
1
a = ( ),
2
1
b = ( + ).
2
Here we have considered modes with an exponential behavior in time. For a particle with M02 <
1/4, it is well known that the flow of energy through the boundary cannot be set to zero and is
given by

,
out flow of energy j = 2T0 = 4 e2 t
(3.18)
=/2
where T is the energymomentum tensor and prime denotes . Both solutions for , however,
can have finite energy for a finite period of time and are hence normalizable. The energy of the
238
mode, E, is given by
/2
E=

d T00 = 2e2 t
.
=/2
(3.19)
/2
It is readily seen that

dE
= j
dt
as it should be. We can now interpret our results as follows; for < 0, if the system starts with a
positive (negative) energy, there will be a net flow of energy out of (into) the space and the mode
dies off in late times finishing with zero energy. However, if > 0 and the system starts with a
positive (negative) energy, the flow of energy will be into (out of) the space and the mode grows
exponentially in time finishing with an infinitely large magnitude for energy in late times which
leads to instability.
The above discussion results in the well-known fact that, to have tachyonic instability coming
from a minimally coupled scalar in an AdS background, we need to violate the Breitenlohner
Freedman bound [28], which for the AdS2 case is exactly M02 < 1/4. For the case when the
four-dimensional cosmological constant vanishes, R1 = R2 and hence M02 = M 2 . If we choose
our non-Abelian gauge group to be SU(2), with a monopole charge n (n > 1), M02 = n/2 (cf.
discussions of Section 2.1) and hence we have tachyonic instability.
Here we briefly address the case R1
= R2 . This happens when we study near horizon limit of
an extremal charged blackhole in a background of non-vanishing four-dimensional cosmological
constant [9,32]. We then have
1
= Q2 ,
R12
1
= Q2 + ,
R22
(3.20)
where Q and the monopole charge n are related as in (3.9). Note that in order to have AdS2 S 2
solution, as is seen from above, || Q2 . The instability condition therefore reads as
n Q2 + 1
> .
2 Q2 4
(3.21)
For > 0 the above condition is always satisfied and we remain with the instability. In fact
negative cosmological constant increases the instability in the sense that it increases the tachyonic mass.
The negative , however, decreases the M02 and hence decreases the instability. It can even
remove the instability. For a given magnetic charge n, if
2
4
gYM
n2
n2 > 4MPl
2n 1
2n + 1
(3.22)
the system becomes stable. Notice that for the negative the (in)stability condition (3.21) is
not the same as the four-dimensional BF bound (as the BF bound analysis is carried out for the
scalars in the AdS background rather than a vector). More detailed analysis of the case with
non-vanishing cosmological constant is postponed to future works [29].
It is remarkable that in the Poincar patch the potential in terms of has exactly the same
expression as the similar case in the flat space (cf. Eq. (3.3)). Note also that, it is AdS in the
239
Poincar patch that is obtained as the direct result of taking the near horizon limit. This may give
a direct realization of the observation we made in the beginning of this section.
As another remark we would like to note that in taking the near horizon limit we are replacing
the localized charges with fluxes (see, e.g., [32]). This in particular implies that the instability
result is not limited to the charges and monopole configuration, but is a generic feature for the
cases we are dealing with non-vanishing non-Abelian fluxes.
4. Instability in EYMD and implications for the attractors
So far we have mainly focused on the EYM theory with a simple gauge group G. We also discussed the EYMH theory and argued how addition of a Higgs field can change the (in)stability
of the background magnetic flux. Here we would like to address implications of addition of the
singlet scalar fields (the dilatons) to EYM. This is the case of relevance to the attractor mechanism [10]. The attractor mechanism simply states that the near horizon geometry of the extremal
charged blackholes only depend on the charges and are independent of the other details of the
theory, and in particular the moduli, if present [10]. In other words, there are (infinitely) many
extremal blackhole solutions which reduce to the same solution in the near horizon geometry and
that the dynamics on the horizon is decoupled from the dynamics of the rest of the space [25].
This result has been used to study the entropy of the extremal blackholes [32] and to identify the
micro-state counting of these blackholes [33].
The attractor behavior was first discussed in the context of supersymmetric blackholes [9],
however, it was then realized that [10] it is a consequence of a weaker requirement than supersymmetry, and extremality suffices for that. Besides the attractor property supersymmetry usually
also implies stability of the background.11 Moreover, for the supersymmetric cases, relying on
supersymmetry, one may perform micro-state counting of a blackhole at weak coupling and
safely extrapolate it to strong coupling regime. The recent realization of the attractor behavior
in non-supersymmetric contexts may bring the hope that other consequences of supersymmetry,
like stability, could also be traced to weaker requirements like extremality. This is the question
we are going to address in this section.
For the cases where we have a single dilaton field and a simple gauge group G, presence of
the dilaton field does not essentially change the equations of motion for the gauge fields and
following the line of computations we have presented here, one can readily show that the same
mode of the gauge field fluctuation which we have shown to be tachyonic remains tachyonic in
the spherically symmetric magnetized solutions of the EYMD theory.
In order to study the implications for the attractor mechanism, recalling that when only the
magnetic charges (fluxes) are turned on one needs to have more than one U (1) factors (e.g., see
[10]), we consider a more general case of the EYMD theory where our gauge group is a product
of several simple factors and also several dilaton fields are present:

1
4
a
bMN
i N j
R + Tr FMN F
fab () g()ij M ,
S = d x g
(4.1)
8GN
where a, b run over 1, . . . , k (k is the number of simple factors in the generic gauge group G),
i are scalar dilaton fields and the range of i, j could be arbitrary, independent of k, gij is
11 Although we do not have a general theorem which ensures that all supersymmetric solutions to (super)gravity theories
are stable, there are positivity theorems which ensure the stability of supersymmetric backgrounds for asymptotically
flat [30] or asymptotically AdS spaces [31].
240
the metric on the moduli space of the dilaton fields. Gauge invariance implies that fab () =
i
eai ab .
The extremal solutions of the above EYMD theory are of relevance to the attractor mechanism [10]. For concreteness let us choose the gauge group to be SU(2) SU(2) (i.e., k = 2) and
take a model with a single dilaton and fab () = ab ea , a, b = 1, 2. The exact attractor solution
of Ref. [10] can be readily embedded in this model. As shown in this reference for 1 2 < 0
there exits an exact attractor solution. Assuming = const, the monopole and blackhole configuration of previous section will correspond to the zeroth order solution of [10]. It is clear that the
analysis of Sections 2.4.22.4.4 of this reference can be carried through verbatim. The instability
in non-Abelian flux backgrounds, however, still remains there, when the background charge is
different from n = 1 for each SU(2). It is also straightforward, along the line of Section 3.2,
to check that the AdS2 S 2 type solutions to the above EYMD theory are also unstable. In an
obvious way one can show that the instability remains for (the extremal) solutions to the above
EYMD (4.1) in its most general form.
As we discussed, the solutions to the EYMD theories which show the attractor behavior,
independently of the value of the dilaton fields at infinity and the moduli, all except for the
minimal charge n = 1 case, suffer from the tachyonic instability and the runaway behavior in
the small fluctuations of some of the components of the gauge fields. Thus before studying the
attractor property, it is necessary to address the instability issue first. For the n = 1 case we expect
to still see the attractor behavior similarly to the Abelian case.
5. Relevance of colored blackholes to string theory
We have discussed that EYMD theories have unstable regular or blackhole solutions. The
instability manifests itself in the modes with imaginary frequency. The EYM and YMD theories
are two specific limits of the EYMD theory. In this part we discuss the relevance of EYMD
theories to string theory.
5.1. Non-Abelian gauge theories and CalabiYau compactifications
Let us start with standard 10d superstrings and first consider the heterotic strings compactified down to 4d. The low energy effective theory obtained in this way is generically a 4d
(super)gravity theory plus some scalars (dilatons) and some gauge fields. In the toroidal compactification, we start with E8 E8 or SO(32) 10d heterotic strings and end up with N = 4
supergravity in 4d with gauge group which is a subgroup of E8 E8 or SO(32), plus 12 other
U (1) vector fields coming from the ten-dimensional metric and two form B-field upon KK reduction.12 In a generic point of the moduli space, where all the 16 E8 E8 or SO(32) Wilson
line moduli are turned on, the gauge group is Higgsed down to U (1)16 [34]. Similarly, if we
start with type I, type II or M theories, upon toroidal compactification to 4d, generally we obtain
a 4d supergravity with U (1)28 gauge group. Therefore, at generic points of moduli spaces of
string/M theory toroidal compactifications non-Abelian gauge groups may be completely bro12 It is possible that in the self T-dual points of the compactification moduli space this Abelian U (1)12 enhances to
non-Abelian SU(2)12 or other larger groups of rank 12 [34].
241
ken, and hence the instability discussed here, does not show up.13 At non-generic points, where
some of the Wilson line moduli are not turned on, we do obtain non-Abelian gauge groups.
The toroidal compactification, although the simplest, is not the most interesting string compactification. A phenomenologically viable string compactification should lead to non-Abelian
GUTs or the Standard Model. For this purpose the CalabiYau (CY) compactification is a better
choice. In this case, for the heterotic or type I case, in the non-generic points of the compactification moduli space, we obtain non-Abelian gauge groups which are subgroups of E8 E8
or SO(32). In these cases we again face the generic instability of the colored blackholes we
discussed here. Even if we are only interested in classical configurations which are free of these
instabilities, the fact that these states exist among the classical solutions of the theory, at quantum
level, can induce instability.
As it is seen from the above discussion effective four-dimensional Abelian theories coming from string theory compactification are generically Higgsed non-Abelian EYM theories (the
Higgs symmetry breaking scale depends on the value of fluxes and the expectation value of the
Wilson line). Therefore, in view of our discussions of previous sections, the seemingly stable
ReissnerNordstrom solution to these theories, although stable in the effective Abelian theory,
once their higher-dimensional origin is noted, can be unstable. Recalling that blackholes of this
type are among the blackholes for which the microstate counting has been performed in string
theory, the microstate counting should be handled with special care.
5.2. Non-Abelian gauge theories in G/H compactifications
The other class of compactifications which leads to non-Abelian gauged supergravities in 4d
are compactification on non-Ricci-flat manifolds, e.g., G/H coset manifolds, G being a compact
group. In these cases the isometries of the compactification manifold, G, appears as the gauge
group in the lower-dimensional theory. Moreover, generically we also obtain a non-vanishing
cosmological constant in the lower-dimensional theory. The value of the cosmological constant
is proportional to the scalar curvature of the compactification manifold, though with the opposite sign. Hence, in these cases we are dealing with (supersymmetric) EYMD- theory. As
discussed in Section 2.2 and to some extent in Section 3.2, appearance of does not generically
cure the instability of non-Abelian solutions. There is, however, the possibility of having stable
non-Abelian solutions. The EYMD- theory will be dealt with in more detail in an upcoming
publication [29].
As the last class we mention the warped compactifications. Here the non-Abelian gauge symmetry can appear in two ways. Either it arises from the bulk gauge fields somehow confined to the
branes [36], or from the internal degrees of freedom on the branes, as in the HoravaWitten scenario [37] or its generalizations and variants. In either cases we face the instability discussed here
and before using these setups for any model building the instability issue should be addressed.
6. Concluding remarks
In this paper we have studied the instability of general spherically symmetric solutions of
four-dimensional EYMD theories with a constant magnetic field strength of generic value on the
13 It is worth noting that closed strings on a torus, when a background magnetic flux larger than some critical value is
turned on, contain tachyonic modes [35]. It has been argued that this tachyonic instability can signal a phase transition in
the system [35].
242
sphere. We showed that the same mode which causes the instability in the gravitational cases
is also present in the non-gravitational case. The instability is due to the fact that the lowest
modes of gauge bosons in the background of a Dirac magnetic monopole on the flat spacetime
is tachyonic and has an exponential growth in time.
We showed that a Dirac magnetic monopole of a generic charge in a non-Abelian gauge theory
causes instability in the fluctuations of the gauge fields. We should stress that, as discussed in
Section 2.1, although this statement is true for a generic charge there exist cases corresponding
to the minimum allowed value of charges that there are no tachyons and are hence stable. For
example, for the SU(2) gauge group, that is n = 1. As discussed in Section 3, these cases remain
stable when gravity is turned on.
One may wonder about the dynamics of the tachyon and ask for a tachyon condensation mechanism with or without gravity. One should note that the tachyon is a feature of the linearized
analysis. Its exponential growth will eventually make the linearized analysis invalid. A tachyon
condensation of some kind may be the final fate of this instability. However, for this an independent analysis is called for.
An interesting outcome of our main result, which was indeed our original motivation for looking into this problem, concerns the attractor mechanism [9,10] in extremal (supersymmetric or
non-supersymmetric) blackholes. These are charged blackhole solutions to EinsteinMaxwell
Dilaton theory. According to the attractor mechanism the near horizon behavior of the blackhole
solution is independent of the value of the dilatonic fields and only depends on the charges defining the solution. In the case of the colored black holes, due to the instability discussed here, the
straightforward application of the attractor mechanism becomes questionable. This result, among
other things, is giving an explicit example of the properties resulting from supersymmetry that
cannot be deduced from extremality.
As we have explicitly shown in Section 3 and is the main theme of this work, the instability in
the presence of the non-Abelian magnetic fluxes is an inherent feature of the non-Abelian fluxes
and not limited to monopoles or colored blackholes. This fact can also be relevant to resolving
the moduli problem in string compactification and removing some of the pieces of the string
landscape [38]. In a sense this instability could be used as another criterion for distinguishing the
swampland from the landscape [39].
Here we mainly focused on the EYM theory in four dimensions. One of the cases which is
of outmost interest is the EYM- theories. As briefly discussed in Section 5.2 these theories are
very relevant to string theory. It is interesting to study the (in)stability of the embedded Abelian,
colored blackhole solutions in the context of 4d gauged supergravities [29]. Among this class the
case of 4d SO(8) gauged supergravity, which is related to M-theory on S 7 and the stability of the
4d AdSRN blackholes are of particular interest. It will also be very interesting to generalize our
analysis of Section 3 to the cases in higher dimensions and check whether the tachyonic modes
still persist. We would expect that the kind of instability present in four dimensions should also
appear in higher dimensions. This expectation should, however, be checked by explicit computations. One of the most interesting cases concerns the charged blackhole configurations of the 5d
SU(4) gauged supergravity, arising from IIB/S 5 . These are the cases relevant to the AdS/CFT
duality [29].
Acknowledgements
We are grateful to Gary Gibbons and the referee for bringing some (old) references on the
subject of colored black holes and their instabilities to our attention. We would like to express
243
our gratitude to the organizers of the Third Crete Regional Conference where this project was
initiated. A.E.M would also like to thank the high energy section of the Abdus-Salam ICTP
where some of this research was carried out.
References
[1] A. Sen, Nucl. Phys. B 440 (1995) 421, hep-th/9411187.
[2] R. Bartnik, J. Mckinnon, Phys. Rev. Lett. 61 (1988) 141;
J.A. Smoller, A.G. Wasserman, Einstein/YangMills equations, Commun. Math. Phys. 151 (1993) 303.
[3] P. Van Nieuwenhuizen, D. Wilkinson, M.J. Perry, Phys. Rev. D 13 (1976) 778;
M.J. Perry, Phys. Lett. B 71 (1977) 234.
[4] J.A. Smoller, A.G. Wasserman, S.T. Yau, Commun. Math. Phys. 154 (1993) 377;
O. Brodbeck, N. Straumann, J. Math. Phys. 34 (1993) 2412;
P. Breitenlohner, P. Forgacs, D. Maison, Commun. Math. Phys. 163 (1994) 141.
[5] G.V. Lavrelashvili, D. Maison, Nucl. Phys. B 410 (1993) 407;
O. Brodbeck, N. Straumann, J. Math. Phys. 37 (1996) 1414, gr-qc/9411058.
[6] M.S. Volkov, D.V. Galtsov, Phys. Rep. 319 (1999) 1, hep-th/9810070.
[7] S. Randjbar-Daemi, A. Salam, J.A. Strathdee, Phys. Lett. B 124 (1983) 345;
S. Randjbar-Daemi, A. Salam, J.A. Strathdee, Phys. Lett. B 144 (1984) 455, Erratum.
[8] G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307.
[9] S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1514, hep-th/9602136.
[10] K. Goldstein, N. Iizuka, R.P. Jena, S.P. Trivedi, Phys. Rev. D 72 (2005) 124021, hep-th/0507096.
[11] S. Randjbar-Daemi, A. Salam, J.A. Strathdee, Nucl. Phys. B 214 (1983) 491.
[12] N.K. Nielsen, P. Olesen, An unstable YangMills field mode, Nucl. Phys. B 144 (1978) 376.
[13] J.R. David, JHEP 0209 (2002) 006, hep-th/0208011.
[14] L. Alvarez-Gaume, S.F. Hassan, Fortschr. Phys. 45 (1997) 159, hep-th/9701069.
[15] K.M. Lee, V.P. Nair, E.J. Weinberg, Phys. Rev. Lett. 68 (1992) 1100, hep-th/9111045.
[16] A.A. Ershov, D.V. Galtsov, Phys. Lett. A 150 (1990) 159.
[17] M.E. Ortiz, Phys. Rev. D 45 (1992) 2586;
P. Breitenlohner, P. Forgacs, D. Maison, Gravitating monopole solutions I, Nucl. Phys. B 383 (1992) 357;
P. Breitenlohner, P. Forgacs, D. Maison, Gravitating monopole solutions II, Nucl. Phys. B 442 (1995) 126, grqc/9412039.
[18] P. Forgacs, S. Reuillon, Phys. Rev. Lett. 95 (2005) 061101, gr-qc/0505007.
[19] M.S. Volkov, O. Brodbeck, G.V. Lavrelashvili, N. Straumann, Phys. Lett. B 349 (1995) 438, hep-th/9502045;
O. Sarbach, M. Heusler, O. Brodbeck, Phys. Rev. D 62 (2000) 084001, gr-qc/9910059;
E. Winstanley, O. Sarbach, Class. Quantum Grav. 19 (2002) 689, gr-qc/0111039;
W.H. Aschbacher, Phys. Rev. D 73 (2006) 024014, gr-qc/0509060.
[20] T. Torii, K.I. Maeda, T. Tachizawa, Phys. Rev. D 52 (1995) 4272, gr-qc/9506018;
M.S. Volkov, N. Straumann, G.V. Lavrelashvili, M. Heusler, O. Brodbeck, Phys. Rev. D 54 (1996) 7243, hepth/9605089;
M.S. Volkov, N. Straumann, G.V. Lavrelashvili, M. Heusler, O. Brodbeck, Phys. Rev. D 54 (1996) 7338, hepth/9605166.
[21] E. Winstanley, Class. Quantum Grav. 16 (1999) 1963, gr-qc/9812064;
J. Bjoraker, Y. Hosotani, Phys. Rev. Lett. 84 (2000) 1853, gr-qc/9906091;
J. Bjoraker, Y. Hosotani, Phys. Rev. D 62 (2000) 043513, hep-th/0002098;
O. Sarbach, E. Winstanley, Class. Quantum Grav. 18 (2001) 2125, gr-qc/0102033.
[22] P. Forgacs, S. Reuillon, Phys. Lett. B 568 (2003) 291, hep-th/0306126;
P. Breitenlohner, D. Maison, G. Lavrelashvili, Class. Quantum Grav. 21 (2004) 1667, gr-qc/0307029;
P. Breitenlohner, P. Forgacs, D. Maison, Commun. Math. Phys. 261 (2006) 569, gr-qc/0412067.
[23] D. Lohiya, Ann. Phys. 141 (1982) 104.
[24] Landau Text Book Series, Quantum Mechanics, p. 73, problem 5.
[25] R. Kallosh, N. Sivanandam, M. Soroush, JHEP 0603 (2006) 060, hep-th/0602005.
[26] K. Goldstein, R.P. Jena, G. Mandal, S.P. Trivedi, JHEP 0602 (2006) 053, hep-th/0512138;
P. Kaura, A. Misra, hep-th/0607132.
[27] J.M. Maldacena, J. Michelson, A. Strominger, JHEP 9902 (1999) 011, hep-th/9812073.
244
[28] P. Breitenlohner, D.Z. Freedman, Phys. Lett. B 115 (1982) 197;

E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[29] A.E. Mosaffa, S. Randjbar-Daemi, M.M. Sheikh-Jabbari, Instabilities in magnetized four- and five-dimensional
gauged supergravity solutions, in preparation.
[30] E. Witten, Commun. Math. Phys. 80 (1981) 381;
J.A. Nester, Phys. Lett. A 83 (1981) 241.
[31] G.W. Gibbons, C.M. Hull, N.P. Warner, Nucl. Phys. B 218 (1983) 173;
P.K. Townsend, Phys. Lett. B 148 (1984) 55.
[32] A. Sen, JHEP 0509 (2005) 038, hep-th/0506177.
[33] A. Dabholkar, A. Sen, S. Trivedi, hep-th/0611143;
D. Astefanesei, K. Goldstein, S. Mahapatra, hep-th/0611140.
[34] J. Polchinski, String Theory, vols. 1 & 2, Cambridge Univ. Press, 1997.
[35] J.G. Russo, A.A. Tseytlin, Nucl. Phys. B 448 (1995) 293, hep-th/9411099.
[36] S. Parameswaran, S. Randjbar-Daemi, A. Salvio, hep-th/0608074.
[37] P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
[38] L. Susskind, hep-th/0302219.
[39] C. Vafa, hep-th/0509212;
H. Ooguri, C. Vafa, hep-th/0605264.
Leptogenesis, dark matter and Higgs

phenomenology at TeV
Pei-Hong Gu a, , Utpal Sarkar b
a The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, 34014 Trieste, Italy
b Physical Research Laboratory, Ahmedabad 380009, India
Received 21 May 2007; received in revised form 3 August 2007; accepted 15 August 2007
Abstract
We propose an interesting model of neutrino masses to realize leptogenesis and dark matter at the TeV
scale. A real scalar is introduced to naturally realize the Majorana masses of the right-handed neutrinos. We
also include a new Higgs doublet that contributes to the dark matter of the universe. The neutrino masses
come from the vacuum expectation value of the triplet Higgs scalar. The right-handed neutrinos are not
constrained by the neutrino masses and hence they could generate leptogenesis at the TeV scale without
subscribing to resonant leptogenesis. In our model, all new particles could be observable at the forthcoming
Large Hardon Collider or the proposed future International Linear Collider.
1. Introduction
Currently many neutrino oscillation experiments [1] have confirmed that neutrinos have tiny
but nonzero masses. This phenomenon is elegantly explained by the seesaw mechanism [2], in
which neutrinos naturally acquire small Majorana [2] or Dirac [3] masses. At the same time,
the observed matterantimatter asymmetry [1] in the universe can be generated via leptogenesis
[312] in these models. In this scheme, the right-handed neutrinos or other virtual particles need
have very large masses to realize a successful leptogenesis if we do not resort to the resonant
effect [6,7] by highly fine tuning.
E-mail address: pgu@ictp.it (P.-H. Gu).

doi:10.1016/j.nuclphysb.2007.08.004
246
P.-H. Gu, U. Sarkar / Nuclear Physics B 789 (2008) 245257
Another big challenge to the standard model is the dark matter [1]. What is the nature of dark
matter? Recently, it has been pointed out [1316] that a new Higgs doublet can be a candidate
for the dark matter if it does not decay into the standard model particles. Although the possibility
of Higgs doublet to be a dark matter candidate was proposed many years back [17], following
the recent proposal [13] a thorough analysis have been carried out [14,18] demonstrating its
consistency with all the recent results. In this interesting scenario, the dark matter is expected to
produce observable signals at the Large Hardon Collider (LHC) [14] and in the GLAST satellite
experiment [18]. Combining this idea, the type-I seesaw and the concept [19] of generation of
the cosmological matterantimatter asymmetry along with the cold dark matter, the author of
[15] successfully unified the leptogenesis and dark matter. However, this scenario need the righthanded neutrinos to be very heavy, around the order of 107 GeV.
In this paper, we propose a new scheme to explain neutrino masses, baryon asymmetry and
dark matter at TeV scale by introducing a Higgs triplet which is responsible for the origin of
neutrino masses, a new Higgs doublet that can be a candidate for the dark matter, and a real
scalar which can generate the Majorana masses of the right-handed neutrinos naturally. A discrete
symmetry ensures that the new Higgs doublet cannot couple to ordinary particles. This same
discrete symmetry will also prevent any connection between the right-handed neutrinos and lefthanded neutrino masses. This allows the right-handed neutrinos to decay at low scale generating
the lepton asymmetry, which will be finally converted to the baryon asymmetry through the
sphaleron processes [20]. This will then explain the observed matterantimatter asymmetry in
the universe, even if the Majorana masses of the right-handed neutrinos are not highly quasidegenerate. In our model, all new particles could be close to the TeV scale and hence should be
observable at the forthcoming LHC or the proposed future International Linear Collider (ILC).
2. The model
We extend the standard model with some new fields. The field content is shown in Table 1, in
which

0
L
L =
(1)
,
=
lL
are the left-handed lepton doublet and Higgs doublet of the standard model, respectively, while

=

(2)
Table 1
The field content in the model. Here L , are the standard model lefthanded lepton doublets and Higgs doublet, is the new Higgs doublet,
R is the right-handed neutrinos, is the real scalar and L is the Higgs
triplet. Here the other standard model fields and the family indices have
been omitted for simplicity
Fields
SU(2)L
U (1)Y
12
12
12
1
0
1
0
3
1
247
is the new Higgs doublet that will be the dark matter candidate, R is the right-handed neutrino,
is the real scalar and
1 +

++
2
L =
(3)
0
1 +
2
is the Higgs triplet. We further introduce a discrete Z4 symmetry, under which the different fields
transform as
L L ,
R iR ,
i,
L L .
(4)
Here the other standard model fields, which are all even under the Z4 , and the family indices
have been omitted for simplicity.
We write down the relevant Lagrangian for the Yukawa interactions,

1
1
c
c
yij Li Rj + gij Ri Rj + fij Li i2 L Lj + h.c. ,
L
(5)
2
2
ij
where yij , gij , fij are all dimensionless. We also display the general scalar potential of , ,
and L ,

2
1
1
V (, , , L ) = 21 2 + 1 4 + 22 + 2
2
4

2

2
+ 23 + 3 + M
Tr L L

2

2
+ 4 Tr L L + 5 Tr L L

+ 1 2 + 2 2 + 3 2 Tr L L

+ 21 + 22

+ 23 Tr L L + 24 L L

+ 25 Tr L L + 26 L L

+ T i2 L + T i2 L + h.c. ,
(6)
where 1,2,3 and have the mass dimension-1, while 1,...,5 , 1,2,3 , 1,...,6 and are all dimen2 is the positive mass-square of the Higgs triplet. Without loss of generality, and
sionless, M
will be conveniently set as real after proper phase rotations.
3. The vacuum expectation values
For 1 > 0 and 21 < 0, we can guarantee that before the electroweak phase transition, the real
scalar acquires a nonzero vacuum expectation value (VEV),

2
u = 1 .
(7)
1
We can then write the field in terms of the real physical field as
+ u,
(8)
248
so that the explicit form of the Yukawa couplings become

1
1
c
c
L yij Li Rj + Mij Ri
Rj + fij Li i2 L Lj
2
2
1
c
+ T i2 L +
T i2 L + gij Ri
Rj
2

2
+ T i2 L + h.c. + M
Tr L L ,
(9)
where we defined,
Mij gij u
and u.
(10)
For convenience, we diagonalize gij gi as well as Mij Mi by redefining Ri and then

simplify the Lagrangian (9) as
1
c i + T i
L yij Li Nj + fij Li
2 L Lj
2 L
2
+
T i2 L + T i2 L + h.c.

1
1
2
+ gi Ni Ni + Mi Ni Ni + M
Tr L L ,
2
2
(11)
with
c
,
Ni Ri + Ri
(12)
being the heavy Majorana neutrinos.

After the electroweak symmetry breaking, we denote the different VEVs as
1 v ,
2
L
1 vL
2
and
u
1 v,
2

and then analyze the potential as a function of these VEVs,
1
1
1
1
V (u , v, v , vL ) = 21 u 2 + 1 u 4 + 22 v 2 + 2 v 4
2
4
2
4
1 2 2 1
1
1
2 2
+ 3 v + 3 v 2 + M
vL + (4 + 5 )vL4
2
4
2
4
1
1
1
+ 1 u 2 v 2 + 2 u 2 v 2 + 3 u 2 vL2
2
2
2
1
1
+ (1 + 2 )v 2 v 2 + (3 + 4 )v 2 vL2
2
2
1
1
1
+ (5 + 6 )v 2 vL2 + v 2 vL + v 2 vL ,
2
2
2
(13)
with u . Using the extremum conditions, 0 = V /u = V /v = V /v = V /vL , we

obtain,
1
0 = 1 u 3 + 21 u + 1 v 2 u + 2 v 2 u + 3 vL2 u + v 2 vL ,
2
2
2
2
2
0 = 2 + 1 u + (1 + 2 )v + (3 + 4 )vL + 2 2vL + 2 v 2 ,
0 = 23 + 2 u 2 + (1 + 2 )v 2 + (5 + 6 )vL2 + 2 2 vL + 3 v 2 ,
(14)
(15)
(16)
2
1
1
0 = v 2 + v 2 + M
+ 3 u 2 + (3 + 4 )v 2
2
2
+ (5 + 6 )v 2 vL + (4 + 5 )vL3 .
For
3 > 0,
23 + 2 u 2 + (1 + 2 )v 2 + (5 + 6 )vL2 + 2 2 vL > 0,
249
(17)
(18)
2 , u 2 ,
the new Higgs doublet gets a zero VEV, i.e., v = 0. We assume < M and v 2
M
and then deduce
v 2
v 2
1
1
vL

2 + u 2 + ( + )v 2
2 + u 2
2 M
2 M
3
3
4
3
1 v 2

2
2 M
for
2
M
3 u 2 .
Subsequently, u and v can be solved,

2 + v2 + v2
2 + 1 v 2
1
3 L
u = 1
1
,
1
1

22 + 1 u 2 + (3 + 4 )vL2 + 2 2vL
2 + 1 u 2
2
,
v=
2
2
for
(19)
(20)
(21)
1 > 0,
2 + 1 v 2 + 3 vL2 < 0,
1
2 > 0,
22 + 1 u 2 + (3 + 4 )vL2 + 2 2vL < 0.
(22)
(23)
We then obtain the masses of resulting physical scalar bosons after the electroweak symmetry
breaking,
2
+ 3 u 2 + (3 + 4 )v 2 ,
M2++ M

1
2
+ 3 u 2 + 3 + 4 v 2 ,
M2+ M
2
2
M20 M
+ 3 u 2 + 3 v,
m2
m20
R
23 + 2 u 2
m
2 + m2 ,
+ 1 v ,
2
(24)
(25)
(26)
(27)
(28)
2 m2 ,
m20 m
(29)
2h m2h ,
m2h1 m
(30)
m2h2
m
2h
+ m2h ,
(31)
with
m
2 23 + 2 u 2 + (1 + 2 )v 2 ,
(32)
250
m2

2
2
v

v ,
2 + u 2
2
M
M
3
(33)
m
2h 1 u 2 + 2 v 2 ,
2
1

m2h 1 u 2 2 v 2 + 412 u 2 v 2 2 .
(34)
(35)
0 are defined by
Here + and R,I
+ ( ) ,

1 0
+ iI0 .
0 R
2
(36)
(37)
In addition, the mass eigenstates h1,2 are the linear combinations of h and , i.e.,
h1 sin + h cos ,
(38)
h2 cos h sin ,
are defined by

1 v+h
,
0
2
where h,
(39)
u + ,
(40)
and the mixing angle is given by

tan 2
21 u v
.
2 v 2 1 u 2
(41)
4. Neutrino masses
The first diagram of Fig. 1 shows the type-II seesaw approach to the generation of the neutrino
masses. It is reasonable to take the scalar cubic coupling less than the triplet mass M in (19).
In consequence, the triplet VEV in (19) is seesaw-suppressed by the ratio of the electroweak
scale v over the heavy mass M . Substantially, the neutrinos naturally obtain the small Majorana
masses,

mII

ij
v 2
1
fij vL fij
.
2
2M
2
(42)
For the zero VEV of new Higgs doublet , we cannot realize the neutrino masses via the
type-I seesaw. However, similar to [15], it is possible to generate the radiative neutrino masses at
one-loop order due to the trilinear scalar interactions in (11). As shown in the second diagram of
Fig. 1, the one-loop process will induce a contribution to the neutrino masses,

m
I ij
m2 0
m2 0

m2 0
m2 0
1
R
R
I
I

=
yik yj k Mk
ln
ln
2
.
2
2

2

2

2
16
m 0 Mk
Mk
m 0 Mk
Mk 2
k
R
(43)
Here Mk uu Mk . For |21 | |1 |v 2 , we have u u and then Mk Mk , so the above formula

can be simplified as
2
2

I
Mk
Mk
1
1
2
2
y
y
ln
ln
m
ij
(44)
m
m
,
ik j k
0
2
2
R
I0
Mk
16 2
m
m
0
0
k
R
251
Fig. 1. The neutrino mass-generation. (a) Is the type-II seesaw. (b) Is the radiative contribution.
2 , we have m2
m
by taking m2 0
Mk2 . Moreover, from (32) and (33), if | |
M
2 and
R,I
then obtain

m
I

ij

2

m2
Mk
v2
1
y
y
yik yj k
1
ln
=
ik j k
2
2
Mk
2Mk
8
m
(45)

2

2
Mk
Mk
1
1 m2
1
ln
=
O
1
ln
.
2
2
2
2
2
4 v
m
4 M
m
2
(46)
for

=O
Note that the above loop-contribution will be absent once the values of and then are taken
to be zero.
5. Baryon asymmetry
We now demonstrate how the observed baryon asymmetry is generated in this model. In the
Lagrangian (11), the lepton number of the left-handed lepton doublets and the Higgs triplet are
1 and 2, respectively, while those of the heavy Majorana neutrinos, the Higgs doublets and
the real scalar are all zero. There are two sources of lepton number violation, one is the trilinear
interaction between the Higgs triplet and the Higgs doublets, the other is the Yukawa couplings
of the heavy Majorana neutrinos to the left-handed lepton doublet and the new Higgs doublet.
Therefore, both the Higgs triplet and the heavy Majorana neutrinos could decay to produce the
lepton asymmetry if their decays are CP-violation and out-of-equilibrium.1
We can obtain the CP asymmetry in the decay of Ni through the interference between the treelevel process and three one-loop diagrams of Fig. 2, in which the first two one-loop diagrams are
the ordinary self-energy and vertex correction involving another heavy Majorana neutrinos, while
the third one-loop diagram is mediated by the Higgs triplet [21]. So it is convenient to divide the
1 Note that there is an equivalent choice of lepton number: L = 1 for and L = 0 for , which makes only the
R
T i2 L term to be lepton number violating. So, the CP asymmetry in the decays of Ni and L can only create
an asymmetry in the numbers of L and an equal and opposite amount of asymmetry in the numbers of . Thus there
is no net lepton number asymmetry at this stage. However, since only the left-handed fields take part in the sphaleron
transitions, only the L asymmetry gets converted to a BL asymmetry before the electroweak phase transition. After
the electroweak phase transition, we are thus left with a baryon asymmetry equivalent to the BL asymmetry generated
from the L asymmetry and an equivalent amount of asymmetry or lepton number asymmetry, which does not affect
the baryon asymmetry of the universe. In the rest of the article we shall not discuss this possibility, since the final amount
of baryon asymmetry comes out to be the same.
252
Fig. 2. The Higgs triplets decay to the leptons at one-loop order.
total CP asymmetry into two independent parts,

c
j [ (Ni Lj ) (Ni Lj )]
i
= iN + i ,
i
(47)
where
i

1
c
Ni Lj + Ni Lj
=
y y ii Mi
8
(48)
is the total decay width of Ni , while

1
1 2
ak
ak
ai
ai
Im
y
y
1
1
+
ln
1
+
+
,
ik
8 (y y)ii
ai
ai
ak
a i ak
k =i

3
ai
1
a
Im
f
y
y
ln
1
+
1
i =
j m ij im
2 (y y)ii
Mi
ai
a
iN =
(49)
(50)
jm
are the contributions of the first two one-loop diagrams and the third one, respectively. Here the
definitions
ai
Mi2
,
2
M1
2
M
M12
(51)
have been adopted.

Furthermore, as shown in Fig. 3, in the decay of L , the tree-level diagram interferes with the
one-loop correction to generate the CP asymmetry,

c
c
ij [ (L Li Lj ) (L Li Lj )]
2

2 /M 2 )
k ln(1 + M
2 ij k (yki ykj fij )M
k
=
,
(52)
2
2
2
Tr(f f )M + 4 + 4
with

c
c
L Li
Lj
ij
+ (L ) + (L )
253

L Li Lj
ij

+ L + L

1 1 2 + 2
Tr f f +
M
=
2
8 4
M
(53)
being the total decay width of L or L .

Note that we have not considered the cases where directly decay to produce the leptons and
anti-leptons through the imaginary Ni or L if m > 2Mi , M + 2m with m and m being
the masses of and , respectively. For simplicity, here we will not discuss these cases.
It is straightforward to see that and i will both be zero for = 0 and then = 0. In
the following, to illustrate how to realize non-resonant TeV leptogenesis, we first focus on the
simple case where iN is the unique source of the CP asymmetry. Note that = 0 for = 0,
accordingly, the one-loop diagram of Fig. 1 is absent and Ni have no possibility for the neutrino
masses, we thus obtain
3 Im[(y y)21k ] M1
16
(y y)11 Mk
k=2,3

3
M 1 M1

+
sin
16 M2 M3
1N
(54)
with being the CP phase. Here we have assumed N1 to be the lightest heavy Majorana neutrinos,
2 , M 2 . The final baryon asymmetry can be given by approximate relation [22]
i.e., M12
M2,3
1
(for K
1),
g
nB
28
YB
(55)

0.31
s
79
(for K 1),
g K(ln K)0.6
Fig. 3. The heavy Majorana neutrinos decay at one-loop order.
254
where the factor 28/79 is the value of B/(B L) and the parameter K is a measure of the
departure from equilibrium and is defined by

1
2 M

1
45
Pl
K
=
y
y
.
11 26 5 g
H (T ) T =M1
M
1
(56)
Here H (T ) = (4 3 g /45) 2 T 2 /MPl is the Hubble constant with the Planck mass MPl
1019 GeV and the relativistic degrees of freedom g 100. For example, inspecting M =
10 TeV, || = 1 GeV and f 106 to (43), we obtain m O(0.1 eV) which is consistent
with the neutrino oscillation experiments. Furthermore, let M1 = 0.1M2,3 = 1 TeV, y 106
and sin = 103 , we drive the sample predictions: K 48 and 1 1.2 105 . In consequence, we arrive at nB /s 1010 as desired.
For = 0 and then , = 0, L and Ni will both contribute to the neutrino masses and the
lepton asymmetry. In the limit of M
Mi , the final lepton asymmetry is expected to mostly
produce by the decay of L . However, because the electroweak gauge scattering should be out
of thermal equilibrium, it is difficult for a successful leptogenesis to lower the mass of L at
TeV scale. Let us consider another possibility that Ni are much lighter than L . In this case,
2 , M 2 and |
2 , N
leptogenesis will be dominated by the decay of Ni . For M12
M2,3
|
M
1
and 1 can be simplified as [10]
1N

I
3 Im[(y y)21k ] M1
3 M1 Im[(m
)j k y1j y1k ] 1

16
8 v 2
(y y)11 Mk
(y y)11
jk
k=2,3
3
8
Imax
M1 m
v2
1
sin ,

II

3 M1 Im[(m )j k y1j y1k ]
3 M1 mII
max

sin ,
2
2
8 v
8
(y y)11
v
(57)
(58)
jk
Imax are the maximal eigenstates of the neutrino

where and are CP phases, mII
max and m
mass matrixes (42) and (45), respectively. Inputting y 107 , M1 = 1 TeV and M2,3 = 10 TeV,
we obtain m
Imax = O(103 eV). Similarly, mII
max = O(0.1 eV) for M = 10 TeV, || = 1 GeV
and f 106 . Under this setup, we deduce 103 by substituting m
= 70 GeV, | | =
N
3
12
10 TeV into (46) and then have 1 2 10
with the maximum CP phase. We also acquire
| | and sin = 0.15. We thus drive the sample predictions: K
1 3 108 for ||
0.5 and 1 1 3 108 . In consequence, we arrive at nB /s 1010 consistent with the
cosmological observations.
6. Dark matter and Higgs phenomenology
0
Since the new Higgs doublet cannot decay into the standard model particles, the neutral R
0
and I can provide the attractive candidates for dark matter [1315]. In particular, to realize dark
0 and 0 should have the mass spectrum [14]:
matter, R
I
m (89) GeV for mL = (6073) GeV,
(59)
m (912) GeV for mL = (7375) GeV.
(60)
255
Here m mN L mL with mL and mN L being the lightest and the next lightest masses between
0 and 0 . Note
R
I
1
m
(mL + mN L ),
2
2 1 2

m m m2 ,
NL
L
2
we thus deduce,

1 |m2 |
1
mL = m
,
2 m
2
m =
|m2 |
m
(61)
(62)
(63)
(64)
In the previous discussions of TeV leptogenesis with = 0, we take M = 10 TeV, || =

1 GeV, ||
= 103 TeV and m
= 70 GeV. It is straightforward to see |m | 25 GeV from (33).
Therefore, we obtain mL 66 GeV and m 9 GeV, which is consistent with the mass spectrum (59).
0 and 0 are expected to be produced in pairs by the standard model gauge bosons W , Z
R
I
0 and a virtual
or and hence can be verified at the LHC. Once produced, will decay into R,I
0 is lighter
W , which becomes a quarkantiquark or leptonantilepton pair. For example, if R
0
than I , the decay chain
+ I0 l + ,
0 +
then I0 R
l l ,
(65)
has 3 charged leptons and large missing energy, and can be compared to the direct decay
0 +
l
+ R
(66)
to extract the masses of the respective particles.

As for the phenomenology of the Higgs triplet at the LHC as well as the ILC, it has been
discussed in [23]. The same-sign dileptons will be the most dominating modes of the ++ . Complementary measurements of |fij | at the ILC by the process e+ e+ (+ ) li lj would allow
us to study the structure of the neutrino mass matrix in detail.
For = O(TeV), which is natural to give the TeV Majorana masses of the right-handed
neutrinos and then realize the TeV leptogenesis, the mixing angle and the splitting between
h1,2 may be large. Furthermore, the couplings of h1,2 to W and Z bosons, quarks and charged
leptons have essentially the same structure as the corresponding Higgs couplings in the standard
model, however, their size is reduced by cos and sin , respectively. In the extreme case = 2 ,
the couplings of the lighter physical boson h1 to quarks and leptons would even vanish. In other
words, this mixing could lead to significant impact on the Higgs searches at the LHC [24,25].
7. Summary
We propose a new model to realize leptogenesis and dark matter at the TeV scale. A real
scalar is introduced to naturally realize the Majorana masses of the right-handed neutrinos. Furthermore, we also consider a new Higgs doublet to provide the attractive candidates for dark
matter. Since the right-handed neutrinos have no responsibility to generate the neutrino masses,
256
which is mostly dominated by the Higgs triplet through the type-II seesaw, they can have large
CP asymmetry at a low scale, such as TeV, to produce the observed matter-antimatter asymmetry
in the universe, even if their Majorana masses are not highly quasi-degenerate. It should be noticed that all new particles are close to the TeV scale and hence should be observable at the LHC
or the ILC.
References
[1] Particle Data Group, W.M. Yao, et al., J. Phys. G 33 (2006) 1.
[2] P. Minkowski, Phys. Lett. B 67 (1977) 421;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified Theory and the Baryon
Number of the Universe, KEK, Tsukuba, 1979, p. 95;
M. Gell-Mann, P. Ramond, R. Slansky, in: F. van Nieuwenhuizen, D. Freedman (Eds.), Supergravity, North-Holland,
Amsterdam, 1979, p. 315;
S.L. Glashow, in: M. Lvy, et al. (Eds.), Quarks and Leptons, Plenum, New York, 1980, p. 707;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912;
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227.
[3] H. Murayama, A. Pierce, Phys. Rev. Lett. 89 (2002) 271601;
B. Thomas, M. Toharia, Phys. Rev. D 73 (2006) 063512;
B. Thomas, M. Toharia, Phys. Rev. D 75 (2007) 013013;
D.G. Cerdeno, A. Dedes, T.E.J. Underwood, JHEP 0609 (2006) 067;
P.H. Gu, H.J. He, JCAP 0612 (2006) 010.
[4] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45;
P. Langacker, R.D. Peccei, T. Yanagida, Mod. Phys. Lett. A 1 (1986) 541.
[5] M.A. Luty, Phys. Rev. D 45 (1992) 455;
R.N. Mohapatra, X. Zhang, Phys. Rev. D 46 (1992) 5331.
[6] M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 345 (1995) 248;
M. Flanz, E.A. Paschos, U. Sarkar, J. Weiss, Phys. Lett. B 389 (1996) 693.
[7] A. Pilaftsis, Phys. Rev. D 56 (1997) 5431;
A. Pilaftsis, T.E.J. Underwood, Nucl. Phys. B 692 (2004) 303.
[8] E. Ma, U. Sarkar, Phys. Rev. Lett. 80 (1998) 5716.
[9] E.Kh. Akhmedov, V.A. Rubakov, A.Yu. Smirnov, Phys. Rev. Lett. 81 (1998) 1359;
K. Dick, M. Lindner, M. Ratz, D. Wright, Phys. Rev. Lett. 84 (2000) 4039.
[10] S. Davidson, A. Ibarra, Phys. Lett. B 535 (2002) 25;
W. Buchmller, P. Di Bari, M. Plmacher, Nucl. Phys. B 665 (2003) 445;
Hambye, G. Senjanovic, Phys. Lett. B 582 (2004) 73;
S. Antusch, S.F. King, Phys. Lett. B 597 (2004) 199;
P. Gu, X.J. Bi, Phys. Rev. D 70 (2004) 063511.
[11] W.L. Guo, Phys. Rev. D 70 (2004) 053009;
G. DAmbrosio, et al., Phys. Lett. B 604 (2004) 199;
S. Antusch, S.F. King, JHEP 0601 (2006) 117;
N. Sahu, U. Sarkar, Phys. Rev. D 74 (2006) 093002.
[12] W. Buchmuller, R.D. Peccei, T. Yanagida, Annu. Rev. Nucl. Part. Sci. 55 (2005) 311.
[13] E. Ma, Phys. Rev. D 73 (2006) 077301.
[14] R. Barbieri, L.J. Hall, V.S. Rychkov, Phys. Rev. D 74 (2006) 015007.
[15] E. Ma, Mod. Phys. Lett. A 21 (2006) 1777.
[16] X. Calmet, J.F. Oliver, Europhys. Lett. 77 (2007) 51002.
[17] E. Ma, S. Pakvasa, S.F. Tuan, Phys. Rev. D 16 (1977) 1568;
N.G. Deshpande, E. Ma, Phys. Rev. D 18 (1978) 2574.
[18] L.L. Honorez, E. Nezri, J.F. Oliver, M.H.G. Tytgat, JCAP 0702 (2007) 028;
M. Gustafsson, E. Lundstrom, L. Bergstrom, J. Edsjo, astro-ph/0703512.
[19] V.A. Kuzmin, hep-ph/9701269.
[20] V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36;
R.N. Mohapatra, X. Zhang, Phys. Rev. D 45 (1992) 2699.
[21] P.J. ODonnell, U. Sarkar, Phys. Rev. D 49 (1994) 2118.
[22] E.W. Kolb, M.S. Turner, The Early Universe, AddisonWesley, Reading, MA, 1990;
H.B. Nielsen, Y. Takanishi, Phys. Lett. B 507 (2001) 241.
[23] E. Ma, M. Raidal, U. Sarkar, Phys. Rev. Lett. 85 (2000) 3769;
E. Ma, M. Raidal, U. Sarkar, Nucl. Phys. B 615 (2001) 313.
[24] W. Buchmller, C. Greub, P. Minkowski, Phys. Lett. B 267 (1991) 395.
[25] O. Bahat-Treidel, Y. Grossman, Y.R. Comments, hep-ph/0611162;
J.R. Espinosa, M. Quirs, hep-ph/0701145.
257
Cluster simulation of relativistic fermions in two

spacetime dimensions
Ulli Wolff
Institut fr Physik, Humboldt Universitt, Newtonstr. 15, 12489 Berlin, Germany
Received 25 July 2007; accepted 20 August 2007
Abstract
For MajoranaWilson lattice fermions in two dimensions we derive a dimer representation. This is equivalent to Gattringers loop representation, but is made exact here on the torus. A subsequent dual mapping
leads to yet another representation in which a highly efficient SwendsenWang type cluster algorithm is
constructed. It includes the possibility of fluctuating boundary conditions. It also allows for improved estimators and makes interesting new observables accessible to Monte Carlo. The algorithm is compatible with
the GrossNeveu as well as an additional Z(2) gauge interaction. In this article numerical demonstrations
are reported for critical free fermions.
1. Introduction
Occasionally in talks or papers about dynamical fermions it is mentionedmore or less as
a jokethat the computer has no data-type Grassmann and one hence can simulate fermions
only via the non-local effective theory after integrating them into the determinant. Of course,
this is plagued by the well-known inefficiencies. In this article, based on Gattringers loop
representation [1], we show that in two spacetime dimensions one actually can get pretty
close to simulating Grassmann numbers.1 We here expand on [1] and its recent numerical
implementation [2] in several ways. First the loop representation is re-derived starting from
Majorana fermions in what we think is a particularly natural way. The new connection includes definite boundary conditions on the torus and does not only work in the thermodyE-mail address: uwolff@physik.hu-berlin.de.
1 Cum grano salis.
doi:10.1016/j.nuclphysb.2007.08.006
U. Wolff / Nuclear Physics B 789 (2008) 258276
259
namic limit as before. In particular, we can then also approach the finite volume continuum
limit. Furthermore we propose a cluster algorithm that is (practically) free of critical slowing down and allows for improved estimators. In this formulation we can simulate fluctuating
boundary conditions which is necessary to allow for fixed (anti)periodic boundary conditions
in the original fermion system. It also makes ratios of partition functions accessible as observables in Monte Carlo simulations. They constitute interesting quantities in the continuum
limit.
The original GrossNeveu model of self-coupled fermions in two dimensions [3] is most
naturally written in terms of N species of Majorana fermions. In the lattice discretization with
Wilson fermions the euclidean action is given by [4]

1
r
g 2 2
S = a2
(1.1)
C + m a
C
.
2
2
8
x
The Grassmann-valued field i (x) has a spin index = 1, 2 and a flavor index i = 1, . . . , N
that we leave implicit. We denote by , , the forward, backward and symmetric nearest
neighbor differences on our cubic T L lattice. The charge conjugation matrix C obeys
C C 1 = = ,
C = C .
(1.2)
For even N each pair of Majorana fermions may be considered as one Dirac fermion

1
1
= (1 + i2 ),
(1.3)
= 1 i2 C
2
2
with its independent , . In the Majorana form the full global symmetry group O(N ) is manifest beside (without Wilson term) the discrete 5 symmetry whose spontaneous breaking was
studied in [3] in the N limit. The model is renormalizable in the strict sense: there is
no other O(N ) invariant scalar 4-fermion interaction term. For N = 2 we have the Thirring
model [5,6], the cases N 3 are expected to be asymptotically free. In the remainder of this paper
we set the Wilson parameter to r = 1 and work in lattice units a = 1. The discrete chiral symmetry 5 of the massless continuum theory is broken by the Wilson term and is only expected
to be recovered in the continuum limit at the critical mass m = mc . On the torus we consider four
conceivable combinations of periodic or antiperiodic boundary conditions in the two directions.
Periodicity angles different from 0, as sometimes used for Dirac fermionswould not lead
to a periodic action density for Majorana fermions. We label the possible boundary conditions
by a bit-vector
,
0 , 1 {0, 1}
(1.4)
where 0 stands for periodic and 1 for antiperiodic boundary conditions in the corresponding
direction.
Often the interaction term is factorized by the introduction of an auxiliary bosonic field. For
us it will be more convenient to think of m m(x) as an x-dependent mass for a while. If Z [m]
is the partition function of one free Majorana fermion with boundary condition in this background field, then the partition function of the interacting theory is written as
2

N

2
g
Z [m] .
Zint = exp
2
2 x m(x)
(1.5)
260
Integrating the fermions in the remaining Gaussian problem yields the Pfaffian

1
Z [m] = Pf C + m .
2
(1.6)
In Appendix A one can find a reminder of the definition of Pfaffians. For even N we may replace
the Pfaffians by the N/2 power of the determinant. In this form and with a factorizing field, the
model can be simulated by standard methods like HMC as carried out in [4]. For larger g this
compute-intensive task became rather difficult due to singularities developing in the operator
under the Pfaffian. Of course, the model itself remains completely well-defined on any finite
lattice. Fermionic and compact bosonic variables are safe in this respect.
As a first step we now introduce the loop representation [1] of Z [m]. It may in fact also be
looked upon as a dimer ensemble similar to those derived in [7] for strong coupling QCD.
2. External field fermion partition function
As a building block for the GrossNeveu models we consider for a single Majorana field the
external field action

1
S=
(2.1)
(x) (x)C(x)
(x, ) (x)CP ()(x
+ ).
2 x
x,
We assume a lattice with T sites in the time direction ( = 0) and L sites in the space direction
( = 1). The variables (x) = 2 + m(x) and (x, ) are external commuting fields. The link
field is introduced for completeness. It will be dropped at some point. The lattice derivatives
in (1.1) combine to Wilson projectors that we define for arbitrary lattice unit vectors n =
1
P (n) = (1 n ),
2
The last identity implies

CP (n) = CP (n).
+ )
= (x + )CP
()(x).
(x)CP ()(x
(2.2)
(2.3)
While the field has torus periodicity , the external fields , are continued periodically to
obtain a periodic action density.
Defining the covariant projecting hop operator
+ )
(H )(x) = (x, )CP ()(x
(2.4)
we may, with the help of (2.3), also write the action in the manifestly antisymmetric short-hand
form

1
1
C
H H
S=
(2.5)
2 x
2 x,
where H is transposed with respect to both spin and space indices yielding

)CP ()(x
).
H (x) = (x ,
(2.6)
Now we are ready for the partition function

Z [, ] = D eS .
(2.7)
The measure is

D = (d1 d2 )(x)
261
(2.8)
and yields

Z [, ] = Pf C
H H
(2.9)
which is a non-local expression in the external fields as in the case of the usual Dirac fermion determinant. In Appendix A we evaluate the Pfaffian for 1 and constant for all four boundary
conditions.
3. Equivalent statistical systems
3.1. Dimer representation
The Grassmannian Boltzmann factor may be expanded,

1 + (x, ) (x)CP ()(x
{1 + 2 1 }
+ )
.
Z = D
x
(3.1)
x,
All fields in the curly bracket are at x and this factor is best considered as part of the measure.
We have here chosen C such that 12 C = 1 2 which just amounts to a phase convention. Note
that the square of the hop-term vanishes due to the one-dimensional projectors. There is only one
linear combination of the two Grassmann numbers contributing from each site which squares to
zero. We now introduce one-bit-valued dimer or bond variables [7] on each link k(x, ) = 0, 1,
whose values are used to organize the expansion as

k(x,)
Z [, ] =
D
(x, ) (x)CP ()(x
{1 + 2 1 }
+ )
. (3.2)
{k(x,)}
x,
As in [7] the goal now is to integrate out the fermions to yield a Boltzmann weight [k] for each
dimer configuration. By asking how the Grassmann integrations can be saturated site by site it is
clear that a non-zero weight only arises if at each site there are either two dimers adjacent from
different links or none at all. In the latter case the integration is saturated by the measure term
and a factor (x) appears for this site. We also call these contributions monomers. Due to the
above constraint the dimers have to form closed non-intersecting and non-backtracking loops.
We choose randomly a starting point and an orientation on such a loop such that along the loop
one visits the sites (x1 , x2 , . . . , xl ). Consecutive sites differ by lattice unit vectors, xi+1 = xi + n i ,
including x1 = xl + n l in the last step. For such a loop the product of bilinears

(x1 )CP (n 1 )(x2 ) (x2 )CP (n 2 )(x3 ) (xl )CP (n l )(x1 )
(3.3)
has to be considered together with the integrations on the l sites involved. Note that here (2.3) is
relevant on the links that are transversed in the negative direction. The trivial key formula is, for
a single site,

d1 d2 = C 1 .
(3.4)
262
Now the expression (3.3) integrates to

X = tr P (n 1 )P (n 2 ) P (n l ) .
(3.5)
Here appears the very important minus sign for a closed fermion loop, well-known for instance
from Feynman diagrams. It is not difficult to see that this result is independent of the starting
point and orientation chosen.
The evaluation of the spin factor follows [8]. Let us introduce eigenspinors of the projectors
P (n i ) = |n i
n i |,
n i |n i = 1,
(3.6)
X =
n 1 |n 2
n 2 |n 3
n l |n 1 .
(3.7)
A spinor is rotated by an angle by the unitary spin matrix R() = exp[(/2)0 1 ]. This allows
us to write
|n i = R( i )|n i+1 ,
i {0, /2}, i = 1, 2, . . . , l
(3.8)
with n l+1 = n 1 . Using
n j |R( i )|n j = cos( i /2)
(3.9)
we get to
X =
n l |n 1
l1

cos( i /2).
(3.10)
i=1
The rotation accumulated in steps (3.8) is

|n 1 = R( l )|n l with =
l
i .
(3.11)
i=1
For closed paths we have =2 and R( l ) = cos()R( l ). For the non-zero

lattice angles, cos( i /2) = 1/ 2. Altogether the final result is
X = ()+1 2Nc /2 ,
(3.12)
where Nc is the number of /2 angles (corners) occurring along the loop and = 0, 1 is
the number of complete rotations the loop makes. The extra minus sign for || = 1 is the one
associated with fermions under 2 -rotations. If we include a non-trivial field then the product
of its Wilson loops for all dimer loops appear in addition in the weight. After this remark we set
1 until further notice.
Although we are on a lattice here, we can define homotopy classes of loops. Two loops are
homotopic to each other if they can be transformed into each other by a sequence of steps, where
dimers are only changed around a single plaquette. We see from the above that X is positive for
all configurations containing only loops that are homotopic to the trivial loop, just a point. The
two minus signs characteristic of fermions compensate each other in this class!
Loops can wind however around the torus in either direction as noted in [2]. A pair of loops
winding around the same direction is still in the trivial homotopy class. An odd number of windings leads however to a new class. This may also happen in both directions at the same time
and hence there are the four classes L00 , L10 , L01 , L11 . In Fig. 1 we show a representative of
each class. They are equilibrium configurations of free fermions at m = 0 and T = L = 10. The
263
Fig. 1. Examples of dimer configurations (solid lines) in the topological classes L00 , L10 , L01 , L11 (from upper left to
lower right). The time direction ( = 0) points to the right. The signs refer to spins on the dual lattice.
meaning of the +/ signs in the plots will become clear later. Only configurations from L00
have a positive weight while in the other cases there is an odd number of closed loops with zero
total rotation angle each of which contributes a factor 1. By introducing antiperiodic boundary conditions in some direction the loops closing around that direction receive yet another sign
without changing the topologically trivial ones.
With the local weight

f (k, x)
[k] =
(3.13)
x
with
(x)
f (k, x) =
1/ 2
if monomer at x,
if 2 dimers in the same direction at x,
if 2 dimers in different directions at x,
else
we now define the positive dimer partition functions

Zk00 [] =
[k],
Zk10 () =

k(x,) L00
[k],
(3.14)
etc.
(3.15)
k(x,) L10
From what was said above the connections

4Zk00 [] = +Z00 [] + Z10 [] + Z01 [] + Z11 [],
(3.16)
4Zk10 [] = Z00 [] + Z10 [] Z01 [] + Z11 [],
(3.17)
264
4Zk01 [] = Z00 [] Z10 [] + Z01 [] + Z11 [],
(3.18)
4Zk11 [] = Z00 [] + Z10 [] + Z01 [] Z11 []
(3.19)
arise, which can be inverted. If we want to realize the boundary conditions = (1, 0) of [4]
(or actually any other definite choice for the fermions) we have to sum over all dimer classes
including negative weight contributions
Z10 [] = Zk00 [] + Zk10 [] Zk01 [] + Zk11 [].
(3.20)
All these relations between partition functions can be turned into relations between expectation values of the scalar fermion density and monomer densities by differentiating with respect
to (x). One example based on (3.20) is

(x)
(x)C(x) 10
2
Z 00
K(x) 00 + Zk10
K(x) 10 Zk01
K(x) 01 + Zk11
K(x) 11
= k
.
(3.21)
Zk00 + Zk10 Zk01 + Zk11
The observable K(x) is one if there is a monomer at x and zero otherwise. With the help of
(x, ) as a source one could establish further relations.
For free fermions ( = 2 + m) in the thermodynamic limit at fixed m > 0 the various Z differ
only by exponentially small amounts and Zk00 dominates among the dimer ensembles. Taking the
finite volume continuum limit (L with Lm fixed, see Appendix A) and in particular for
m = 0 this is not so. In the latter case we have an exact zero fermion mode for = (0, 0) and
Z00 = 0 holds. This implies
Zk00 = Zk10 + Zk01 + Zk11
for m = 0.
(3.22)
3.2. Spin representation

In this subsection we transform the dimer system to yet another representation by Ising spins.
This will allow us to design a global cluster algorithm. A clue that this may be possible is given
by the idea that a natural way to manage and modify the closed loops in the dimer formulation
is to consider them as boundaries of domains of up-spins surrounded by down-spins (Peierls
contours).
The spins that we introduce live on the lattice dual to the one carrying the fermionic and the
dimer variables. Its sites labelled by underlined x are dual to plaquettes of the original lattice and
are imagined to be located at their centers. Analogously, the sites of the original lattice are dual
to plaquettes in the new one. Links (x, ) and (x, ) are dual to each other if they cross, see
Fig. 2. The idea is now to put an Ising field s(x) on the dual lattice and to identify configurations

1 if s(x)s(x + )
= 1,
k(x, ) =
(3.23)
0 if s(x)s(x + )
= +1.
In other words, dimers are located where nearest neighbor spins on the dual lattice are antiparallel. In a first stage we restrict ourselves to the class L00 of dimer configurations.
We first prove that for each admissible dimer configuration there are exactly two spin fields
obeying (3.23) that differ by a global spin-flip. In a first step we define a Z(2) lattice gauge field2
2 This field has nothing to do with the earlier .
265
Fig. 2. Illustration of the labelling of the original and the dual lattice.
on the dual lattice in terms of the dimers on the original lattice

+1 if k(x, ) = 0,
( x, ) =
1 if k(x, ) = 1.
(3.24)
Because of the constraints on k this gauge field is unity when multiplied around any plaquette
(on the dual lattice). As we restrict ourselves to L00 here, also loops around the torus are unity for
this gauge field. Thus it a pure gauge on the torus with a periodic gauge function. The spin-field
is this gauge function that we construct now. We choose a site y (the origin, for example) and set
s(y) = +1. Now this value is parallel-transported with to all other sites, for instance along a
maximal tree rooted at y. Due to the absence of curvature in , the result is path-independent,
consistent and unique. Starting from s(y) = 1 we obtain the other configuration associated with
{k(x, )}. The signs in Fig. 1 are just these spins.
While we now have exactly two spin fields for each admissible dimer configuration in L00 ,
not all conceivable spin fields are reached in this way. Obviously, spin configurations, that on any
plaquette look like

s4 s3
+
+
(3.25)
=
,
s1 s2
+
+
do not occur in the image, where we here just gave simple labels to the four spins. They would
correspond to crossing loops not allowed by the original Grassmann variables. If this is however excluded on all plaquettes, then we can reconstruct admissible {k(x, )} configurations
from {s(x)}. The total Boltzmann factor in the spin representation is now a big product with one
factor for each plaquette. These weights w are given in Table 1 which lists only 8 of the 16 configurations and is completed by using w(s1 , s2 , s3 , s4 ) = w(s1 , s2 , s3 , s4 ). The monomer
strength (x) is taken here at the site x of the original lattice sitting at the center of the dual
plaquette considered.
The derivation of this representation resembles the construction of the dual formulation for
generalized Ising models [9]. We summarize it (giving the ordinary self-dual two-dimensional
Ising model as an example in brackets): One first introduces new variables living on the bonds
making up the Hamiltonian of the original theory (a link field). Then the original spins are
summed over producing a constraint in the new variables (vanishing plaquettes of the links interpreted on the dual lattice). This constraint is then solved on the dual lattice (links given as a
pure gauge by a site field). The extension of the concept here is that we change from Grassmann
266
Table 1
Plaquette weights depending on the spin configuration
s1
s2
s3
s4
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
(x)
1/2
1/2
1/2
1/ 2
1
1
0
elements to bosonic variables, have an additional constraint (3.25) to fulfill, and that there can be
minus signs. One could talk of FermiBose- or super-duality.
The plaquette weight can also be written in terms of pairwise nearest neighbor bondinteractions of the form [writing ij si sj 12 (1 + si sj )]

w(s1 , s2 , s3 , s4 ) = p[12 + 23 + 34 + 41 ] + q[12 34 + 14 23 ]

+ r[12 14 + 21 23 + 32 34 + 41 43 ] ,
(3.26)
with the x-dependence in p, q, r suppressed. To match Table 1, the following equations have to
hold
(x) = 4p + 2q + 4r,
1
= 2p + r,
2
1 = 2p + q.
The solution of this system is
1
q = 1 + r.
(3.27)
2
We remark here that all coefficients are positive for 0 m 2 2 2.83. In the free case, this
clearly covers the relevant range of bare masses.
If we now include dimer configurations in the other sectors by using (3.23), the only difference
is that the resulting s(x) is antiperiodic in the direction orthogonal to those where dimer loops
run around the torus. Thus L10 corresponds to spin-fields antiperiodic in space, L01 to those in
time and L11 to spins with both directions antiperiodic. The mechanism here is that antiperiodic
boundary conditions of the spins force an interface into their configurations which leads to the
non-trivial dimer loop topology. Introducing further partition functions Zs for the spin ensembles
we relate
1
1
1
1
Zk10 = Zs01 ,
Zk01 = Zs10 ,
Zk11 = Zs11 .
Zk00 = Zs00 ,
(3.28)
2
2
2
2
The factor 1/2 cancels the global spin-flip symmetry. Again, by differentiation, we may relate
expectation values, where the presence (absence) of a monomer yielding K(x) = 1(0) in the spin
language translates into maximally polarized plaquettes where all four spins are parallel,

s(x).
K(x) = 4,|M(x)| , M(x) =
(3.29)
m
1
r = ( 2) = ,
4
4
1
r
p= ,
2 2 2
x around x
267
4. Simulation algorithms
4.1. Local algorithms
A local algorithm to simulate any one of the above dimer ensembles with all weights taken
positive was recently described and tested by Gattringer et al. [2]. The simplest case to consider is
a free Majorana fermion of mass m with (x) = 2 + m. In the updates one actually performs only
changes that are local in the homotopy-sense by proposing dimer-flips k(x, ) 1 k(x, )
around plaquettes. The move is accepted with the Metropolis probability corresponding to the
ratio of in (3.13) for the new and the old configuration, which is a locally computable quantity.
Of course, it vanishes, if the new configuration would violate one of the constraints. This update
stays in the homotopy class fixed by the starting configuration (see Fig. 1) and is ergodic within
it. Thus the various ensembles can be simulated which correspond to combinations of periodic
and antiperiodic Pfaffians. In [2] it was demonstrated that already such simulations are vastly
more efficient than HMC type simulations.
The entirely equivalent update in the Ising form consists of local spin-flips. The Metropolis
decision in this case depends on the eight nearest and next-to-nearest (diagonal) neighbors that
share plaquettes with the spin in focus. The numerical efficiency in both forms is very similar.
4.2. Cluster algorithm
The plaquette interaction in (3.26) is now written as a superposition of 10 different terms,
schematically given by
w(s1 , s2 , s3 , s4 ) =
10
Pi i (s1 , s2 , s3 , s4 )
(4.1)
i=1
with Pi {p, q, r} and (s1 , s2 , s3 , s4 ) {0, 1}. In complete analogy to [10] we introduce now
ten-valued bond variables b(x) with each value corresponding to one of these terms to obtain

Zs =
(4.2)
Pb(x) b(x) (s1 , s2 , s3 , s4 ).
{b(x),s(x)} plaq
To avoid a too clumsy notation here again s1 , s2 , s3 , s4 are spins around each of the plaquettes.
The celebrated trick of [10] consists of Monte Carlo sampling both the b and the s variables.
As the first part of an update cycle one chooses new b at fixed s by a local heatbath procedure. Of course, in general, the choice is between less than 10 possibilities, as some of the
b(x) (s1 , s2 , s3 , s4 ) vanish. Then, for given bonds b, any of the 2T L spin configurations has either
weight zero or a constant nonzero weight depending on the constraints given by the product of all
factors b(x) . Just as in the standard Ising model case we now construct the percolation clusters
defined by the active bonds in all {b(x) }. By flipping the spins in each cluster as a whole with
probability 1/2, we sample one of the allowed and equally weighted spin configurations. The
overall procedure amounts to a global independent sampling of spins (at fixed bonds) and will be
numerically demonstrated to almost eliminate critical slowing in Section 5.
At the stage of selecting a new spin field one may also construct improved cluster estimators.
This is achieved, if, for some observable, one is able to analytically average over all conceivable
spin-assignments of which only one is taken as the next configuration.
268
While the above procedure is analogous to the well-known SwendsenWang algorithm [10]
we could also study a single cluster variant [11]: one spin is chosen at random, and then only
the one cluster connected to it is constructed by investigating the plaquette terms touched in the
growth process until it stops. Then spins on this cluster are always flipped. This may well be even
more efficient as large clusters are preferred.
We end this subsection with a remark on the global spin flip symmetry. At first one may
think that it is a (slightly) annoying redundancy in the new representation. However, it is in fact
essential to be able to grow clusters whose energy (action) is associated with the surface (in our
case the loops) and not with the bulk. Loosely speaking, as one grows a (single) cluster, there is
always an energetic way out by flipping the whole lattice. Of course, if this is all that happens,
that algorithm will not be efficient. The auxiliary percolation problem allows to find non-trivial
clusters.
4.3. Fluctuating boundary conditions
In (3.20) we saw that it is desirable to also be able to simulate enlarged ensembles where
one sums beside configurations of spins also over several possible boundary conditions. If in
conventional simulations one proposes a change of the boundary conditions at fixed spins, one
generates energy (action) proportional to T or L and the proposal will practically always be
rejected. It was noticed however in [12] that with cluster algorithms the situation can be different.
In the step where we pick new spins at fixed bonds the search among possible equally weighted
new configurations can be enlarged to also include changed boundary conditions. If we label
them by again (for the spins now) then the four possible become a dynamical variable.
In [12] these changes were introduced in a single-cluster/Metropolis spirit, which would also be
possiblein fact less involvedhere. In view of the future construction of improved estimators
we stick however to the many-cluster view and design now a correspondingly generalized cluster
algorithm. It consists of the following steps:
We throw bonds on the links as discussed before.
We determine by some percolation algorithm (e.g. tree search) the independent spin clusters
connected by bonds but ignore two layers of links such that the torus is cut open. We take
{(x, 0)|x0 =T 1 } and {(x, 1)|x1 =L1 }. We call these clusters preclusters, their connectivity is
determined in the interior. Each of them carries a unique cluster label as a result.
Now the remaining links are examined as far as they have been activated. We call these
bonds clamps. They have the effect of sewing up (some of) the preclusters. This is done by
the pointer technique described in [13]. We may visualize the process as a graph with the
preclusters as blobs, some of which get connected by lines.
In this process la [13] one can detect when closed loops in the graph are formed. We set
one of four types of flags whenever a loop is closed. They distinguish whether an odd or an
even number of temporal or spatial clamps are met around the loop. We end with flags f00 ,
f10 , f01 , f11 each being zero or one.
Boundary conditions are only compatible with this loop structure if for all occurring loops
with flag up (f = 1) the condition 0 + 1 = 0 mod 2 holds. Of course, = (0, 0) is
always allowed.
Among the compatible (1, 2 or 4 values) one is chosen with equal probability.
Now flips for all preclusters are determined and executed. Connected components of the
graph flip together, but preclusters within these components can flip relative to each other if
269
the boundary condition has changed. The above construction guarantees that the orientations
thus propagated do not depend on the path that is taken on the graph.
Although rather short and compact in the end this is not a trivial code to write. It is helpful to
organize it under a geometric point of view focussing on parallel transport between preclusters
with a Z(2) group, where the boundary conditions are gauge variables on the clamps. Of course,
as we can solve the free fermion ensembles exactly (Appendix A) and easily get high accuracy
with cluster simulations, many significant checks by short simulations (taking seconds) were
available. A very good monitor for debugging at every stage is to set traps for the occurrence of
illegal plaquettes (3.25). An alternative strategy to move boundary conditions would be to turn
the loop structure of the above graph into a system of linear equations in the Galois field of two
elements (addition isomorphic to logical xor). Following [14] and [15] this can be solved by
Gauss elimination.
The above scheme contains some nested pointer operations. One could be worried in principle
whether the execution time grows more than linearly with the lattice volume. In practice there was
found to be absolutely no problem of this kind. This is in fact the same for cluster simulations of
the standard Ising model using the algorithm of [13]. In our case the problem is even less severe
as the number of clamps is smaller than T + L, not proportional to the volume.
4.4. Negative mass
For free fermions one could content oneself with the parameter range m 0 = mc . On the
other hand all results in Appendix A can be taken at arbitrary m. After all, the partition function
on the finite lattice is just a polynomial. When we later come back to the interacting theory, it
will also turn out, that negative mass fermions will be required because mc < 0 due to renormalization. The local algorithm [2] works for negative m, a sign problem only arises if = 2 + m
changes sign. The bond probabilities (3.27) however restrict the cluster algorithm so far to m 0.
Luckily, there is an alternative decomposition of the plaquette interaction into bonds that comes
to rescue when m(x) is negative.
The r-term in (3.26) is replaced by
r
[12 14 + 21 23 + 32 34 + 41 43 + 12 14 + 21 23 + 32 34 + 41 43 ],
2
where we introduced also antibonds ij 1 ij . The two other terms are unchanged but the coefficients are now p,
q.
Using 12 14 = 12 12 14 , etc., one sees that the new weight coincides
with the old one if we identify r = r and p = p + r . The matching equations are now
(x) = 4p + 2q,
1
= 2p + r ,
2
1 = 2p + q + 2r ,
solved by
1
m
r = (2 ) =
,
4
4
1
1
r
p = ,
q = 1 r .
2 2 2
2
Now all weights are positive for m 2(2 2 ) 1.17.
(4.3)
270
The presence of antibonds is compatible with the cluster search including fluctuating boundary conditions. With an m(x) that changes sign over one lattice, one actually decomposes some
plaquettes with (3.27) and others with (4.3). One must however not make the mistake to think
of the preclusters as ferromagnetic (Weiss) domains, they contain in general both up and down
spins. This is why we took care to talk about distributing flips to clusters rather than assigning
new spin orientations to them as whole.
5. Numerical applications
We now report on numerical experiments. In this first publication on the method we stick to
free fermions and the observable K(x). It corresponds to the scalar fermion density and is mainly
used to diagnose the algorithm. Hence, with the results of Appendix A, every computed mean
value is known exactly and was verified to be reproduced by the Monte Carlo simulations within
errors. We do not plot results for K. They are too boring: errors not visible on the graph and
exact results agreeing with the data within 1 and occasionally up to around 2 sigma.
For the algorithm, the non-interacting case does not seem to be fundamentally different from
the interacting GrossNeveu model. All details necessary for this extension are given, but the
numerical implementation is deferred to a future investigation. What remains to be seen is how
the correlation between monomers of different flavor influences the Monte Carlo dynamics at
stronger coupling.
5.1. Critical slowing
We performed a series of simulations of one species of free Majorana fermions at the critical
value m = 0. In this case, the only infrared scale is given by the system size T = L = 8 . . . 128.
We simulated the trivial ensembles corresponding to the loop class L00 . Results are summarized
in Table 2. Each run with the local algorithm, passing through the lattice in lexicographic order,
consists of 106 sweeps of which a small fraction3 is discarded for thermalization. The autocorrelation time int,K has been defined and measured as described in [16].
As one expects for local algorithms we see a steeply rising autocorrelation time hinting at a
dynamical exponent not too far from twowe have no ambition here to determine it precisely
which would be very costly. One notices, that the error (at fixed sweep number) is almost independent of L. This means the variance just compensates the growing autocorrelation time and
decays roughly proportionally to 1/T L. This is in fact implied by scaling and the canonical diTable 2
Monomer density and its integrated autocorrelation time for local and cluster simulations at m = 0 and lattice sizes T = L
Exact
Local
int,K
Cluster
K
int,K
8
16
32
64
128
0.80738
0.78914
0.77951
0.77466
0.77223
0.8080(5)
0.7885(4)
0.7795(4)
0.7742(5)
0.7721(4)
5.13(7)
14.4(3)
44.0(1.6)
187(17)
444(58)
0.80774(32)
0.78932(21)
0.77949(13)
0.77464(8)
0.77221(5)
1.55(2)
1.90(2)
2.29(3)
2.84(4)
3.50(5)
3 We discard the first (T /16)2 1000 sweeps in the local runs.
271
mension of the (connected) 2-point function of the scalar density. Although the integral over the
autocorrelation function at L = 128 does not look too unconvincing, one may suspect that our
number for int,K may only be a lower bound for this case.
The cluster simulations in the last columns consist of 0.6 106 sweeps which in our implementation takes about the same time as the local runs on a single PC. We see small slowly rising
autocorrelation times. From the two largest lattices one would estimate an effective dynamical
exponent zeff 0.30, which is a typical value for cluster algorithms. In total only about 15 CPU
hours went into these demonstrations. All codes have been programmed in MATLAB and the
update routine has about 100 lines (50 without fluctuating boundary conditions).
The cluster algorithm performs very similarly also in the other topological sectors. We did the
T = L = 128 runs with spin boundary conditions = (0, 1) (L10 ) and found int,K = 2.68(4)
and = (1, 1) (L11 ) with int,K = 2.29(3), even shorter than the trivial sector.
The next series of runs to be reported is on T = L = 128 lattices at several positive and
negative masses. Again each data point is produced by 0.6 106 sweeps. These simulations included fluctuating boundary conditions. Recorded observables were the monomer density K, the
boundary conditions and the topological flags f . From these data the distribution of boundary conditions can be deduced and we checked their correctness. As an example we consider
here (3.21) and compute for the right-hand side
S10 =

2 + m
K 2
K,(1,0)
(x)C(x) 10 =
,
2
1 2
,(1,0)
(5.1)
where on the right-hand side the Ising ensemble with fluctuating boundary conditions is meant.
The result at m = 0 is S10 = 0.76987(6) with the exact value being 0.769800361 . . . . Errors for
this combination of observables are estimated as discussed in [16], where the definition of int,S10
from the fluctuations relevant for this quantity can be found. We show these autocorrelation
times in Fig. 3. There seems to be a steep rise by about one unit close to m = 0. The second
plot shows a better resolution of its vicinity. All these numbers stay comfortably small. The
combination measured here has no serious sign problem. One could however construct positive
cluster estimators for numerator and denominator.
A simple example for the use of an improved estimator exploiting the topological information
can be given by the two observables with equal mean

1
1
,(1,0) = (f10 ,f01 ,f11 ),(0,0,0) + (f10 ,f01 ,f11 ),(0,1,0) .

(5.2)
4
2
The fractions on the right hand are the part of admissible boundary conditions given by = (1, 0).
Flag positions not tagged here do not allow this value at all. At m = 0 we find in our run for the
left-hand side 0.18650 (66) [int = 0.852(6)] and for the other mean 0.18650 (37) [int = 1.38(1)]
while the exact answer is 0.186455866 . . . . Of course, the two estimates are strongly correlated.
It is clear that the left estimate is more stochastic using the actually picked boundary conditions
in the run and int is hence smaller. This pattern is typical for cluster estimators with the reduced
variance usually overcompensating this effect.
5.2. Four fermion interaction
To simulate the interacting GrossNeveu model (1.1) we represent each factor Z in (1.5) by
spin-ensembles. Their are summed over independently with weights depending on the desired
boundary conditions for the fermions. At each site x an overall configuration holds 0 n(x) N
272
Fig. 3. Integrated autocorrelation time for the scalar density (5.1) corresponding to fermion boundary conditions antiperiodic in one direction and periodic in the other at T = L = 128.
monomers. The only thing that the interaction does is to change the corresponding weight into a
factor
2 2
[n/2]

n!
g
n
=
g 2j (2 + m)n2j = c(n, m, g).
(2
+
m)
exp
(5.3)
2 m2
2j j !(n 2j )!
j =0
We update one of the spin flavors at a time. Let us introduce the number n(x)
of monomers in
the other momentarily frozen spins. Then in our update the effective monomer weight (or local
mass) has to be taken as
|n (x) =
c(n(x)
+ 1, m, g)
.
c(n(x),
m, g)
(5.4)
The first few terms are

|0 = 2 + m,
|1 = 2 + m +
|2 = 2 + m + 2g 2
g2
,
2+m
2+m
,
(2 + m)2 + g 2
....
(5.5)
In interacting theories the mass of Wilson fermions undergoes additive renormalization. Both
in the Thirring model (N = 2) and in the higher N GrossNeveu model perturbative and nonperturbative calculations [4] as well as large N approximations yield negative values for the
critical mass mc (g 2 ) close to which one wants to simulate. This is why it was crucial to extend
the cluster algorithm to also accommodate < 2.
273
5.3. Additional gauge coupling

An amusing exercise is to add to the Majorana flavor group a color Z(2) gauge interaction.
We describe it here only very briefly. We now make use of the gauge links in (2.1) and consider

N
Zcolor =
(5.6)
e p p Zs [m, ]
(x,)
with the plaquette field p made from the links (x, ) = 1. The self-interaction can always
be added, changing only the monomer weights. We suppress it here. Each dimer loop receives as
an additional factor the Wilson loop made of . By Stokes theorem they can be replaced by a
product of p either on all plaquettes where a + spin resides or on those where the negative spins
sit, see Fig. 1. Let us choose +. In the case of antiperiodic boundary conditions additional loops
around the torus appear where the torus closes (where the clamps were). As a two-dimensional
gauge theory is rather trivial the sum can be carried out. We need to know how many plaquettes
get tiled with p an odd number of times. Let us introduce the composite flavor spin-field
S(x) = s (1) (x)s (2) (x) s (N ) (x).
Then
M =
S(x),
(5.7)
(5.8)
contain this information and the extra weight from summing over is

22T V cosh()M+ sinh()M + cosh()M sinh()M+ .
This is equivalent to a magnetic field
1
H = ln tanh()
(5.9)
2
coupled to S and fluctuating in sign. In addition we only get contributions if an even number
of flavors is antiperiodic in each direction separately (due to Z(2) confinement). The fluctuating
magnetic field can presumably be included in the cluster simulation without problems. As we
update a given flavor, each spin can in addition bond to one exterior (phantom) spin.
6. Conclusion and outlook
It seems that with the new cluster algorithm the GrossNeveu model (and the Thirring model)
are wide open for high precision simulation. Further observables, in particular cluster estimators,
remain to be constructed. The O(N ) Noether currents may be accessible, for example. Also ratios
of partition functions like Zs00 /Zs10 , etc., are expected to have a continuum limit and may serve
as renormalized finite volume couplings. Majorana fermions are prominent in supersymmetry.
Maybe some studies on two-dimensional supersymmetry become possible with simulations very
close to the continuum limit.
An obvious question that every reader will have is if any of this carries over to higher dimension and/or more complicated gauge interactions. Let us first caution here: fermions in
one space dimension are very special. This has manifested itself in other so-called FermiBose
equivalences in two dimensions. At the heart of this in operator language is the fact that the
274
JordanWigner [17] transformation transforms anticommuting to commuting degrees of freedom without generating non-localities. This has no obvious generalization to higher dimension
(see however [18]). The fact that we find positive weights for all topologically trivial loops in
a way seems to be the euclidean counterpart. Also that Majorana fermions are in some sense
equivalent to Ising spins it not new, of course, see [19,20]. The main achievement here is that
we simulate in a standard (lattice) euclidean fermion formulation and can get really critical. In
higher dimension a dimer representation for fermions can probably be constructed along similar
lines, but the weight will be sign-fluctuating in a more essential fashion. The slight hope may
be that one could be able to handle this sign problem with cluster estimators. In addition, the
coupling to gauge fields contributing fluctuating Wilson loops is an open problem. There are
ongoing efforts to simulate discrete models dual to non-Abelian gauge theories (spin foam) (see
[21] and references therein). This may be an interesting view on gauge theories in the context
of the approach to fermions developed here. In any case, the goal is attractive enough to warrant
further thought.
Acknowledgements
We would like to acknowledge discussions about the GrossNeveu model on the lattice with
Francesco Knechtli, Bjrn Leder, Rainer Sommer and most of all Tomasz Korzec. We also thank
the Deutsche Forschungsgemeinschaft (DFG) for support in the framework of SFB Transregio 9.
Appendix A. Exact Majorana partition functions
The Pfaffian is defined for an antisymmetric matrix Aij of even size, i, j = 1, . . . , 2n,

1
1 1
Pf(A) = d 2n e 2 A = n i1 i2 ...i2n Ai1 i2 Ai3 i4 Ai2n1 i2n ,
(A.1)
2 n!
where we integrate over 2n Grassmann variables and a sign convention for the measure is implied. By a change of variables F one sees the well known identity

Pf F AF = det(F ) Pf(A),
(A.2)
and by squaring the integral in (A.1),

2
Pf(A) = det(A)
(A.3)
follows.
We are interested in A = C( + m 12 ) with boundary conditions . The determinant
immediately follows by Fourier diagonalization

p 2 + M(p)2
det(A) =
(A.4)
with

p=

2
2
(n0 + 0 /2),
(n1 + 1 /2) ,
T
L
n0 = 0, . . . , T 1, n1 = 0, . . . , L 1
(A.5)
p = 2 sin(p /2).
(A.6)
and
p = sin(p ),
1
M(p) = m + p 2 ,
2
275
In this article we also want to know the relative signs of Pf(A) for different boundary conditions
and possibly negative m. This information is lost in the determinant and we proceed differently.
We define the unitary Fourier transformation matrix
1 ipx
e
Fxp =
TL
(A.7)
which is augmented trivially to include a unit matrix in spin space. Then, using (A.2) we get
det(F ) Pf(A) = Pf(A)
(A.8)

A qp = C p + M(p) p+q,0 .
(A.9)
with
The factor det(F ) is just a phase that we fix later.

The matrix A consists of antisymmetric blocks where momenta p, p get paired. If they are
different (modulo 2 ) such a block contributes one factor (p 2 + M(p)2 ) to the Pfaffian.
For simplicity we now restrict our discussion to both T and L even. Then all momenta get
paired non-trivially except p = (0, 0), (, 0), (0, ), (, ) in the all periodic case = (0, 0).
We may summarize this result by

sign[m(m + 4)] for = (0, 0),
p 2 + M(p)2
Pf (A) =
(A.10)
1
other .
p
Possible extra phase factors can be excluded by considering the limit m . Note that the
double periodic Pfaffian changes sign at m = 0.
In this article we need for comparison the scalar condensate in ensembles that in fermion
language have the partition function

Z [m] =
(A.11)
c()Z [m].
It is given by
1
1
1
C c =
ln(Z ) =
2
T L m
2
c()z(, m)
C
c()z(, m)
(A.12)
with
Z [m]
z(, m) =

Z [m]
(A.13)
and
1
M(p)
1
.
C =
2
T L p p 2 + M 2 (p)
(A.14)
All these exact results can be easily evaluated for any finite lattice that is simulated. For =
(0, 0), m 0 the product z
C (0,0) remains finite but requires precaution numerically.
276
Table 3
Symanzik expansion coefficients for ratios of partition functions at different boundary conditions in the finite volume
continuum limit at fixed = mL
(1, 0)
d0
1
1
1
1
(0, 0)
(1, 0)
(0, 0)
(1, 0)
0.3135575596
d2
d4
0.219897
1.1
d0
d1
d2
0.12681663
0.27632955
0.16991195
0.37023294
0.048092
0.012935
0.08631
0.030368
0.108
0.135
0.183
0.285
The continuum limit of z can presumably be computed analytically. We here content ourselves
with the numerical construction of their Symanzik expansion in a few cases. We set T = L
(aspect ratio one) and

dk (, )Lk
z(, m)|mL=
(A.15)
k0
and compile some values in Table 3.

The analysis of the asymptotic series was carried out as described in Appendix D in [22].
The missing in the table can be computed from those given. Digits are quoted such that there
is at most an uncertainty of one in the last digit. For = 0 odd corrections vanish.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
C. Gattringer, Nucl. Phys. B 543 (1999) 533, hep-lat/9811014.

C. Gattringer, V. Hermann, M. Limmer, arXiv: 0704.2277 [hep-lat].
D.J. Gross, A. Neveu, Phys. Rev. D 10 (1974) 3235.
T. Korzec, U. Wolff, PoS LAT2006 (2006) 218, hep-lat/0609022.
W.E. Thirring, Ann. Phys. 3 (1958) 91.
S.R. Coleman, Phys. Rev. D 11 (1975) 2088.
P. Rossi, U. Wolff, Nucl. Phys. B 248 (1984) 105.
I.O. Stamatescu, Phys. Rev. D 25 (1982) 1130.
F.J. Wegner, J. Math. Phys. 12 (1971) 2259.
R.H. Swendsen, J.S. Wang, Phys. Rev. Lett. 58 (1987) 86.
U. Wolff, Phys. Rev. Lett. 62 (1989) 361.
M. Hasenbusch, J. Phys. (Paris) I 3 (1993) 753, hep-lat/9209016.
J. Hoshen, R. Kopelman, Phys. Rev. B 14 (1976) 3438.
R.C. Brower, S. Huang, Phys. Rev. D 41 (1990) 708.
B. Bunk, Fermion Algorithms, World Scientific, Jlich, 1991.
ALPHA Collaboration, U. Wolff, Comput. Phys. Commun. 156 (2004) 143, hep-lat/0306017.
P. Jordan, E.P. Wigner, Z. Phys. 47 (1928) 631.
E.H. Fradkin, M. Srednicki, L. Susskind, Phys. Rev. D 21 (1980) 2885.
T.D. Schultz, D.C. Mattis, E.H. Lieb, Rev. Mod. Phys. 36 (1964) 856.
J.B. Kogut, Rev. Mod. Phys. 51 (1979) 659.
J.W. Cherrington, D. Christensen, I. Khavkine, arXiv: 0705.2629 [hep-lat].
ALPHA Collaboration, A. Bode, P. Weisz, U. Wolff, Nucl. Phys. B 576 (2000) 517, hep-lat/9911018.
Three-loop chromomagnetic interaction in HQET

A.G. Grozin a,b , P. Marquard c , J.H. Piclum a,c , M. Steinhauser c,
a Department of Physics, University of Alberta, Edmonton, Alberta T6G 2G7, Canada
b Budker Institute of Nuclear Physics, Novosibirsk 630090, Russia
c Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, 76128 Karlsruhe, Germany
Received 11 July 2007; accepted 20 August 2007

Available online 4 September 2007
Abstract
We compute the three-loop QCD corrections to the quark chromomagnetic moment and thus obtain
the matching coefficient and the anomalous dimension of the chromomagnetic interaction in HQET. As a
byproduct we obtain the three-loop corrections to the quark anomalous magnetic moment.
PACS: 12.39.Hg; 12.38.Bx
Keywords: Heavy quark effective theory; Radiative corrections
1. Introduction
We consider Quantum Chromodynamics (QCD) with nl light flavours and one heavy
flavour Q. The interaction of a single heavy quark having momentum mQ v + k (mQ is the
on-shell mass and v 2 = 1) with gluons and light quarks in the situation when the residual momentum k mQ (and momenta of light fields are also small) is described by the Heavy Quark
Effective Theory (HQET) Lagrangian [1,2]
L = Q v iv DQv +

1
1
Ok + Ccm ()Ocm () + O
,
2mQ
m2Q
E-mail address: matthias.steinhauser@uka.de (M. Steinhauser).

doi:10.1016/j.nuclphysb.2007.08.012
(1)
278
A.G. Grozin et al. / Nuclear Physics B 789 (2008) 277293
where v/ Qv = Qv is the HQET quark field and D denotes the covariant derivative (see the books
[3,4] for more details). The kinetic energy operator
2
v D
Ok = Q
Qv ,
D = D v(v D),
(2)
does not renormalize and its coefficient is equal to 1, to all orders, due to reparametrization
invariance [5]. The chromomagnetic interaction operator is defined as
1
Ocm = Q
v G Qv ,
2
(3)
where G = gs Ga ta is the gluon field strength tensor, ta is a colour matrix, s = gs2 /(4) the
strong coupling and = 2i [ , ]. It is responsible for the violation of the heavy-quark spin
symmetry and thus, e.g., for the BB mass splitting. The coefficient Ccm () is obtained by
matching the scattering amplitudes of an on-shell heavy quark in an external chromomagnetic
field, expanded in the momentum transfer q up to the linear term, in the full theory (QCD) and
the effective theory (HQET). If all flavours except Q are massless, all loop corrections in HQET
vanish. It is most convenient to calculate the QCD scattering amplitude using the background
field method [6].
The chromomagnetic interaction coefficient Ccm () has been calculated at one-loop order in
Ref. [1]. The one-loop anomalous dimension cm of the chromomagnetic operator (3) follows
from its dependence; it has also been found in Ref. [2]. The two-loop anomalous dimension
has been obtained in Refs. [7,8], and the two-loop matching coefficient Ccm () in Ref. [8]. All
orders of perturbation theory in the large-0 limit were summed in Ref. [9]. The effect of a nonzero charm quark mass mc on the bottom-quark chromomagnetic interaction at two loops has
been investigated in Ref. [10]. In this paper (cf. Section 3) we calculate cm , as well as Ccm ()
at three loops, provided that all light flavours are massless. Using these results, we obtain the
next-to-next-to-leading perturbative correction to the ratio
R=
m2B m2B
m2D m2D
(4)
which we discuss in Section 4.

The mass difference between the vector and pseudo-scalar B meson is also often studied with
lattice gauge theory simulations where non-perturbative results for the operator matrix element
can be obtained (see, e.g., Ref. [11] for an introduction to lattice HQET and Ref. [12] for a
recent study). In principle there are two possibilities to make contact with the experimentally
measured result: the matching can be performed perturbatively and non-pertubatively. In the first
case the perturbatively computed n-loop matching coefficient and the (n + 1)-loop result for
the corresponding anomalous dimension have to be combined with the non-perturbative lattice
results. Currently this is done for n = 1 which according to Ref. [12] induces an uncertainty of
about 4%. With the new results of this paper this uncertainty can be significantly reduced. This
topic is discussed in Section 5.
We also investigate the heavy-quark magnetic moments. They have not yet been measured
experimentally, however, for the bottom and the lighter quarks there are upper limits from LEP1
data [13]. For the bottom quark this limit is close to the Standard Model (SM) prediction including the two-loop QCD correction [14]. Thus, a more precise measurement at a future linear
collider should be able to determine the bottom-quark anomalous magnetic moment and probe
possible deviations from the SM.
279
Fig. 1. Sample diagrams contributing to the quark chromomagnetic moment. Solid, curly and dotted lines denote quarks,
gluons and ghosts, respectively. represents the coupling of the background field. In the closed quark loops all flavours
have to be considered.
The top-quark magnetic moment has not been measured so far. However, such a measurement
would be very interesting since the top-quark couplings to photons or Z bosons are very sensitive
to contributions from physics beyond the SM. Having this in mind, it is mandatory to have precise
SM predictions for these couplings. In Section 6 we provide the three-loop QCD corrections to
the coupling of the photon to heavy quarks.
2. The calculation
To calculate the chromomagnetic moment we have to consider the quarkanti-quarkgluon
vertex in the background-field formalism in QCD. Sample diagrams are depicted in Fig. 1. When
both the quark and anti-quark are on the (renormalized) mass shell and have physical polariza
tions, the vertex a = ta can be decomposed into two form factors,

= F1 q 2

i
q F2 q 2 ,
2mQ
(5)
where q = p1 p2 is the gluon momentum and p1 and p2 are the momenta of the quark and
anti-quark, respectively.
The anomalous chromomagnetic moment is given by c = Z2OS F2 (0), where Z2OS is the quark
wave function renormalization constant in the on-shell scheme. The total quark colour charge is
given by Z2OS F1 (0) = 1. Thus, F1 (0) is the inverse of the on-shell wave function renormalization
constant, which has been calculated to three-loops in Ref. [15] (see also Ref. [16]). Therefore,
the calculation of F1 (0) provides a strong check on the correctness of our result.
In order to extract the form factors, we use projection operators. They are conveniently obtained by introducing the momentum p = (p1 + p2 )/2, since p q = 0. With this definition we
have

F1 q 2 =
1
2(d 2)(q 2 4m2Q )

4mQ (d 1)
Tr ( p
/ 1 + mQ ) +
p
+
m
)
(
p
/
,
2
Q
q 2 4m2Q
(6)
280

F2 q 2 =
2m2Q
(d 2)q 2 (q 2 4m2Q )

4m2Q + (d 2)q 2
+
m
)
(
p
/
.
p
Tr ( p
/ 1 + mQ ) +
2
Q
mQ (q 2 4m2Q )
(7)
Since the projector for F2 develops a pole for q 2 = 0, we cannot set q 2 = 0 from the beginning.
Instead, we expand in q and keep all terms which are at most quadratic in q. In the final result
the limit q 2 = 0 can be taken. Due to the expansion in q all occurring integrals are on-shell
propagator-type integrals.
All Feynman diagrams are generated with QGRAF [17] and the various topologies are identified with the help of q2e and exp [18,19]. In a next step the reduction of the various functions to
so-called master integrals has to be achieved. For this step we use the so-called Laporta method
[20,21] which reduces the three-loop integrals to 19 master integrals. We use the implementation of Laportas algorithm in the program Crusher [22]. It is written in C++ and uses GiNaC
[23] for simple manipulations like taking derivatives of polynomial quantities. In the practical
implementation of the Laporta algorithm one of the most time-consuming operations is the simplification of the coefficients appearing in front of the individual integrals. This task is performed
with the help of Fermat [24] where a special interface has been used (see Ref. [25]). The main
features of the implementation are the automated generation of the integration-by-parts (IBP)
identities [26] and a complete symmetrization of the diagrams. The master integrals are known
from [15] (see also comments in Ref. [16]). To calculate the colour factors, we have used the
program described in Ref. [27].1
The calculation is performed for an arbitrary gauge parameter in order to use its cancellation
as a check. However, at three-loop level the expressions for the individual diagrams become very
big. Thus we discard all terms with more than linear dependence. This also concerns the factors
1/(1 ) appearing in the vertex of a background field with two quantum gluons, which does
not appear in the usual formulation of QCD. If this is done, our final result is gauge-parameter
independent up to terms which are quadratic in . Furthermore, our calculation of F1 reproduces
Z2OS including the gauge-dependent terms [15,16].
3. Chromomagnetic moment
The renormalized scattering amplitude of an on-shell heavy quark with initial momentum
p1 = mQ v and final momentum p2 = mQ v q in an external gluon field is given by the vertex
(5) sandwiched between u(p
2 ) and u(p1 ) and multiplied by Z2OS . We expand this amplitude in q
up to linear terms, and re-express (relativistic) QCD spinors via (non-relativistic) HQET spinors:

2
k/
k
u(mQ v + k) = 1 +
(8)
+O
uv (k).
2mQ
m2Q
Then the QCD scattering amplitude reads

q
i
q (1 + c ) ta uv (0),
u v (q) v
2mQ 2mQ
1 We thank Philipp Kant for providing his interface for the program of Ref. [27].
(9)
281
(we have used u(p

2 )/
q u(p1 ) = 0). It must be reproduced by the HQET Lagrangian of Eq. (1). If
all flavours except Q are massless, all loop corrections vanish,2 and the scattering amplitude is
given by the Born approximation:

q
i
0
u v (q) v
(10)
q Ccm
ta uv (0),
2mQ 2mQ
0 = Z 1 ()C () is the bare chromomagnetic interaction coefficient, and Z ()
where Ccm
cm
cm
cm
is the MS renormalization constant of the chromomagnetic operator given in Eq. (3). Note that
both scattering amplitudes (9) and (10) are renormalized and hence ultraviolet-finite, however,
both have infrared divergences. These divergences are the same, because HQET has been constructed to reproduce the infrared behaviour of QCD. Vanishing loop correction in HQET have
0
ultraviolet and infrared divergences which cancel each other. The ultraviolet divergences of Ccm
1 (); the infrared ones match those of 1 + (cf. Eq. (9)).
are removed by Zcm
c
In order to find the chromomagnetic interaction coefficient Ccm () in the HQET Lagrangian
(1), we calculate the anomalous chromomagnetic moment c = Z2OS F2 (0) and re-express it in
(n )
terms of s l (), where the superscript denotes the number of active flavours. Then

Ccm () = Zcm s(nl ) () 1 + c s(nl ) () ,
(11)
where nl = nf 1 is the number of light-quark flavours, which are considered to be massless in

our calculation, and nf is the total number of quark flavours.
The coupling constants s(nl +1) () in QCD (with nl + 1 flavours) and s(nl ) () in HQET (with
nl flavours) are related by [28]
(nl +1)
()
(nl )

()
1 2
s () 2
1 2
1
=
TF L +
+
L +
3
6
36

1 2
1
1 3
L + L 3 2 + O 3
+
18
36
9
(nl )

3

s ()
5
1
15
2
1
+
TF
L+
CF +
L
CA + TF L2
4
16
12
9
9

1 2 15
5 2 4
43
1 2 31
5 2
+
L + L+ +
CF +
L L+
+
CA
4
8
48
32
12
9
144
108

1 3
1
+
L + 2 L TF + O 2 + O s4 ,
9
54
(nl )
(12)
where L = ln(2 /m2Q ) and mQ is the on-shell mass of the heavy quark. CF = (Nc2 1)/(2Nc )
and CA = Nc are the eigenvalues of the quadratic Casimir operators of the fundamental and
adjoint representation for the SU(Nc ) colour group, respectively. In the case of QCD we have
Nc = 3 and TF = 1/2. n denotes Riemanns zeta function with integer argument n.
2 We imply the use of dimensional regularization.
282
The ultraviolet divergences contained in Zcm can be transformed into an anomalous dimension
which is given by
(nl ) 2

(n )
d ln Zcm s l 1
s
13
17
CA
=
CA +
CA TF n l
cm =
d ln
2
36
36

(nl ) 3
d abcd dAabcd
899
1
1
s
3 +
CA3 + 2 F
+
8
1728
2
CF N F

1
65
1
49
1
2
3 +
C
3
CA CF + CA TF nl TF nl
2
216 A
2
96
36
abcd d abcd

d
2
F
2 F
nl + O s4 ,
(13)
3
CF N F
where dFabcd and dAabcd are the symmetrized traces of four generators in the fundamental and adjoint representation, respectively (for SU(Nc ), dFabcd dFabcd = (Nc2 1)(Nc4 6Nc2 + 18)/(96Nc2 ),
dFabcd dAabcd = Nc (Nc2 1)(Nc2 + 6)/48). NF = Nc is the dimension of the fundamental representation. The two-loop result agrees with [7,8], and the n2l part of the three-loop one with [9].
Our result for Ccm reads

(n )
s l (mQ ) 1
1
1
L+
CA + CF
Ccm () = 1 +
4
2
2
2

(nl )
3
5 2 31
1 2
s (mQ )
C2
ln 2 + 3 +
+
2
4
12
16 F

1
1
1 2 269
1 2
+
L + ln 2 3 + +
CF CA
8
12
8
12
144

1 2 13
1
17 2 805
25
1 2
+ L + L + ln 2 3
+
CA2 CF TF nl
12
36
12
8
144
432
36

1 2 13
1 2 119
1 2 299
+
L L
CA TF nl + +
C F TF
24
72
36
432
3
36

(nl )

5 2 149
s (mQ ) 3 (3)
+
ccm + O s4 .
(14)

CA TF +
72
216
The two-loop corrections were already calculated in Ref. [8].3 It is convenient to decompose the
three-loop contribution in terms of the different colour structures as
abcd d abcd
(3)
F A dF
A
ccm
= XF F F CF3 + XF F A CF2 CA + XF AA CF CA2 + XAAA CA3 + Xdd
CF N F

+ XF F l CF2 + XF Al CF CA + XAAl CA2 TF nl + (XF ll CF + XAll CA )TF2 n2l

+ (XF lh CF + XAlh CA )TF2 nl + XF F h CF2 + XF Ah CF CA + XAAh CA2 TF
abcd dFabcd
l
h dF
+ (XF hh CF + XAhh CA )TF2 + Xsin
nl + Xsin
.
CF N F
(15)
3 Note that in Ref. [8] the term 2 C T is not correct, since the O() term in the one-loop decoupling relation for
s
A F
(12) has not been taken into account.
283
Our results for the individual terms read

20
5 4
5
22
235
103 2
a4 +
ln 2 2 ln2 2 2 ln 2
5 +
3
3
18
18
3
24
72
139 4 241
23
101
+
3 + 2
,
(16)
24
6
64
2160
35
1
3
5
31
35 4
13
XF F A = 2 ln 2 + 3 + 2
L a4
ln 2 + 2 ln2 2
8
16
48
64
3
72
9
101 2
115
17
209 4 847
9767 2 2803
ln 2 +
5 2 3

3 +

,
24
6
9
2880
96
1728
288
(17)

191
1 2
1 2
1
1 2 337
ln 2 3 + +
L+
a4
XF AA = L +
24
48
32
48
576
18
191 4
169 2 2
1745 2
265
161 2
ln 2
ln 2 +
ln 2
5 +
3
+
432
216
432
16
72
491 4 2951
17375 2 122971
(18)
+
3
+
,
+
10368
288
3456
10368

19 3
497 2
1 2
1
17 2 2917
L
L +
ln 2 + 3
+
L
XAAA =
432
1728
48
32
576
3456
8
1
1
317 2
925
653 2
a4 ln4 2 2 ln2 2
ln 2 +
5
3
3
9
36
864
192
864
17 4 6079
1585 2 1302797

3 +
+
,
(19)
810
1728
1296
186624
1
40
5
20
91
FA
Xdd
= 2 L a4 ln4 2 + 2 ln2 2 + 2 ln 2 105
4
3
9
9
12
97 2
7 4 73
151 2
5
(20)
3
,
+ 3 +
72
270
24
27
18
8
1
2
5
11 4
79
125
33 2 +
,
XF F l = a4 ln4 2 2 ln2 2 + 2 ln 2 +
(21)
3
9
9
3
216
54
32

4
5
1
299
1 4
1
XF Al = L2 +
3
L + a4 +
ln 2 + 2 ln2 2
96
4
576
9
54
27
1 2
23 4 2
23 2 88351
(22)
ln 2
+ 3
,
108
3240
3
72
10368

4
35 3 235 2
1 2
715
1 4
1
XAAl =
L +
L
3 +
+
L + a4 +
ln 2
864
864
4
144
1728
9
54
1
89 2
1 4 101
35 2 236801
(23)
ln 2 +

3 +

,
+ 2 ln2 2
27
216
180
432
864
46656
1
317
XF ll = 2 +
(24)
,
27
324
7
1 3
13 2
1
25 2 3535
XAll =
(25)
L
L L + 3 +
+
,
108
216
72
54
324
5832
1
61
,
XF lh = 2
(26)
27
162
11 2 167
XAlh =
(27)
+
,
108
162
XF F F =
284
32
4
4
16
4 4 263
a4 + ln4 2 2 ln2 2 2 ln 2 +

3
3
9
9
9
135
72
11 2 2027
+
+
,
216
162
1
119
5 4
5
83
L 10a4
ln 2 + 2 ln2 2 + 2 ln 2
XF Ah = 2 +
12
144
12
12
27
25
1
101 4 6937
22241 2 8447
5 + 2 3

3
+
,
8 1440
864
19440
864
24
5 2 149
1 4
1
1211 2

L + a4 +
ln 2 2 ln2 2 +
ln 2
XAAh =
288
864
24
24
432
65
65 2
53 4 4423
283429 2 71965
5 +
3 +
+
3

,
144
432
5184
1728
155520
10368
8
4 2 943
XF hh = 3

,
3
135
324
4
1 2 487
XAhh = 3 +
+
,
9
270
972
1
29 4
44
2
l
Xsin
= 2L +
33 2 + ,
3
270
27
3
2 4
2 2 2
5
5
41 4
h
2
Xsin = 16a4 + ln 2 ln 2 24 ln 2 + 5 2 3
3
3
6
18
540
4
931 2 5
3 +
+ ,
3
54
9
XF F h =
(28)
(29)
(30)
(31)
(32)
(33)
(34)
with a4 = Li4 (1/2). The n2l part agrees with the result obtained in Ref. [9].
Substituting numerical values of the constants, we obtain, for the physical SU(3) colour group,
2
Ccm (mQ ) = 1 + 0.6897s(nl ) (mQ ) + (2.2186 0.1938nl ) s(nl ) (mQ )

3

+ 11.079 1.7490nl + 0.0513n2l s(nl ) (mQ ) + O s4
2
= 1 + 0.6897s(nl ) (mQ ) + (1.16260 0.9786) s(nl ) (mQ )

3

+ 1.846802 + 0.33700 3.8137 s(nl ) (mQ ) + O s4 ,
(35)
with 0 = (11CA /3 4TF nl /3)/4. The first-order 1/0 result [9] contains the highest powers
(n )
of 0 in each term. For nl = 4, for example, the coefficient of [s l (mQ )]2 is 2.4221 0.9786 =
(n )
1.4435, and that of [s l (mQ )]3 is 8.0155 + 0.7020 3.8137 = 4.9039; the large-0 approximation of Ref. [9], which only includes the first terms in these sums, overestimates these two
coefficients by 68% and 63%, correspondingly.
For the numerical evaluation of Ccm (mQ ) and cm , we use the values mc = 1.6 GeV,
mb = 4.7 GeV and mt = 175 GeV. The number of light-quark flavours nl is three, four and
(n )
five for the charm, bottom and top quark, respectively. To evaluate s l (mQ ), defined with
(5)
nl active flavours, from s (mZ ) = 0.118, we use the program RunDec [29] and obtain
(3)
(4)
(5)
s (mc ) = 0.3348, s (mb ) = 0.2163 and s (mt ) = 0.1074, which leads to
Ccm (mc ) = 1 + 0.2309 + 0.1835 + 0.2362 = 1.6506,
(36)
Ccm (mb ) = 1 + 0.1492 + 0.0676 + 0.0497 = 1.2664,
(37)
Ccm (mt ) = 1 + 0.0741 + 0.0144 + 0.0045 = 1.0930.
(38)
285
In the case of the charm quark, we see that the perturbative series does not converge which
is probably connected to the relatively light scale of 1.6 GeV at which the strong coupling is
evaluated. Thus, it seems that there are potentially large non-perturbative corrections. While the
situation is better in the case of the bottom quark, the corrections are still very large. The threeloop correction amounts to about 30% of the one-loop contribution. For the top quark, we find
that our new term contributes about 6% of the one-loop correction leading to a fairly reliable
prediction of Ccm (mt ).
The numerical evaluation of the anomalous dimension (13) gives
2
cm = 0.4775s(nl ) + (0.4306 0.0549nl ) s(nl )

3

+ 0.8823 0.1472 nl 0.0007n2l s(nl ) + O s4 .
(39)
For the individual quark flavours this leads to
cm (mc ) = 0.1599 + 0.0298 + 0.0163 = 0.2060,
cm (mb ) = 0.1033 + 0.0099 + 0.0029 = 0.1160,
cm (mt ) = 0.0513 + 0.0018 + 0.0002 = 0.0533.
(40)
Thus, as far as the anomalous dimension is concerned the convergence behaviour is acceptable
even for the charm quark.
These observations are in good agreement with the analysis of Ref. [9]. The chromomagnetic
interaction coefficient Ccm (mQ ) has the leading renormalon singularity (namely, a branch point)
at the Borel parameter u = 1/2, quite close to the origin; it leads to a very fast growth of coefficients of the perturbative series, n!(0 /2)n . It also means that the leading non-perturbative
correction is only suppressed by the first power of 1/mQ , and is thus important, especially for
the charm quark. On the other hand, the perturbative series for the anomalous dimension has a
finite radius of convergence.
4. Application: Heavy-meson mass splittings
The most prominent physical effect caused by the chromomagnetic interaction is the mass
splittings of hadronic doublets which are degenerate at mQ = due to the heavy-quark spin
symmetry. For example, for the bottom mesons B and B one has [30],

4 (4)
QCD
2
2
2
mB mB = Ccm ()G(4) () + O
(41)
,
3
mb
where the index (4) means that we are considering nl = 4 flavour HQET, and 2G(4) () is
the matrix element of Ocm () (cf. Eq. (3)) over the ground-state meson. It is most natural to
choose = mb in Eq. (41), because then Ccm contains no large logarithms. A similar formula
(3)
can be written down for the D mesons where Ccm () and 2G(3) () appear in the corresponding
expression for m2D m2D . The running of 2G(nl ) () is governed by the anomalous dimension
given in Eq. (13). Furthermore, it is necessary to relate the matrix elements in the two theories
via the following decoupling relation

(4)
(4)

s (mc ) 2
s (mc ) 3
2G(4) (mc ) = 2G(3) (mc ) 1 + z2
(42)
+ z3
+ O s4 ,
286
with z2 = 71CA TF /432 [31,4]. We introduce the unknown4 coefficient z3 since it appears in
estimates of higher order effects which are presented below.
In the formulation with resummed logarithms one combines for consistency Eq. (42) with the
two-loop result for Ccm and the three-loop anomalous dimension. This leads to the ratio R of
Eq. (4) to the next-to-next-to-leading (NNL) order approximation. For later use we extend the
formalism to NNNL order where we obtain

(4)
s (mb )
R = x 0 /(20 ) 1 + r1 (x 1)

(4)
2

r
s (mb ) 2
+ r20 + r21 x 2 1 + 1 (x 1)2
2

3

r
+ r30 + r31 (x 1) x 2 + x + 1 + 1 (x 1)3 + r1 r20 (x 1)
6
(4)

3

s (mb )
QCD
+ r1 r21 (x 1)2 (x + 1)
(43)
+ O s4 ,
,
mb,c
with
x=
s(4) (mc )
(4)
s (mb )
(1)
r1 = ccm

0 1 1
,
20 0 0
(2)
(2)
(nl = 4) ccm
(nl = 3) + z2 ,
r20 = ccm
2

(1) 2
)
(ccm
0
1
2 1 1 2
+
+ z2 +
+
,
2
40
0 0 0 0
0

(3)
(3)
(1) (2)
(2)
ccm (nl = 4) ccm
(nl = 4) ccm
(nl = 3) ccm
(nl = 3) + d2 + z3 ,
r30 = ccm
(2)
(nl = 3) +
r21 = ccm
(1) 3

(ccm
)
(3)
(1) (2)
r31 = ccm
ccm (nl = 3) d2
(nl = 3) + ccm
+ z3
3
2
3

0
1 1 3
1 2
1
3 1 2 2 1
+
+
+
2
+
+
,
60
0 0 0 0 0
0 0 0
0 0
0
(44)
(n)
ccm denotes the coefficient of (s (mQ )/)n in Ccm ( = mQ ). The terms z2 and z3 stem from
the decoupling of the matrix element and are introduced in Eq. (42) and d2 = [(2/9)CA
(15/16)CF ]TF stems from the decoupling of s (cf. Eq. (12)). n are the coefficients of
(s /)n+1 in the anomalous dimension and the coefficients of the function are used in the
form 0 = (11CA /3 4TF nl /3)/4; see Refs. [34,35] for the remaining i .
Both for n and n nl = 4 active flavours have to be chosen. Let us mention that the next-toleading (NL) order result of Eq. (43) has been obtained in Ref. [7]. Inserting the numerical values
given above and displaying the contributions from the individual orders separately, we find
R = 0.8517 0.0696 0.0908 + [0.1285] +
= 0.6914 + [0.1285] + ,
4 z can be calculated using 3-loop HQET integrals considered in [32,33].
3
(45)
287
where the ellipses denote terms of higher order and power corrections. The term in square brackets is our estimate of the fourth order contribution, where we assume that the four-loop coefficient
of the anomalous dimension is negligible. This is justified by the rapid convergence of cm in the
case of the bottom quark as can be seen in Eq. (40). Furthermore, we set the unknown coefficient
z3 of Eq. (42) to zero which is a good approximation since it enters with a small coefficient.
The experimental value is Rexp = 0.88 [36] with a negligible uncertainty. The NNL correction
amounts to 10% of the LO contribution, however, it is larger than the NL one. Furthermore it is
negative and thus increases the difference of the perturbative result and the experimental value.
The estimated third-order correction is even larger than the NL and NNL one and contributes
also with a negative sign. This indicates that the QCD /mc correction may be quite substantial.
It is interesting to consider the quantity R also without performing the resummation of the
logarithms. In this way the three-loop result for the coefficient can be incorporated in a consistent way. The starting point is Eq. (4) where quantities defined for nl = 4 are present in the
numerator and the ones defined for nl = 3 in the denominator. Using Eq. (42) for the decoupling
of the matrix element and running from = mc to = mb cancels 2G(4) . Afterwards, we can
replace s(3) (mc ) by s(4) (mb ), using decoupling and renormalization group running, and perform
(4)
a consistent expansion of R in s (mb ). As a result we obtain

(4)

s (mb )
1
(1)
+ 1 l
+ 0 0 0 l 2 20 ccm
R = 1 0 l
2

(4)
s (mb ) 2
(2)
(2)
+ ccm
(nl = 4) ccm
(nl = 3) + z2

4
1
(1)
+ 0 02 + 0 0 02 l 3 + (0 20 ) 20 ccm
+ 1 1 0 l 2
6
3
(1) 2
(2)

(2)
(2)
+ 0 ccm
(nl = 3) ccm
(nl = 4) 40 ccm
(nl = 3) + 20 ccm

(1)
(3)
(3)
21 ccm
+ (40 0 )z2 2 l + ccm
(nl = 4) ccm
(nl = 3)

(4)

s (mb ) 3
(1) (2)
(2)
ccm
(46)
ccm (nl = 4) ccm
(nl = 3) + d2 + z3
+ O s4 ,
where l = ln(mb /mc ). In our numerical evaluation we set the decoupling coefficient z3 to zero.
Since similar three-loop decoupling effects are small we expect the same in our case. Further(3)
more, note that ccm is numerically rather large. Inserting the numerical values yields
R = 1 0.1113 0.0780 0.0755 + = 0.7352 + .
(47)
Comparing Eqs. (45) and (47), we find that the convergence of R without resummation behaves
slightly better. However, the coefficients of the perturbative series are still large.
5. Matching coefficient and renormalization group invariant quark mass
In this section we would like to discuss the result of the matching coefficient in the form which
is often used in lattice simulations of the mass difference m2B m2B . In doing so we follow the
procedure outlined in Ref. [37].
In lattice simulations one usually determines the renormalization group invariant (RGI) matrix element of the operator Ocm which has to be multiplied by the corresponding matching
288
coefficient. For its derivation one considers in a first step

mag

)
s (m

0
mag
mag 0
s (m
) 20
ds
exp
,
Cmag = 20
2
20
s
(48)
Q (m
) is the scale invariant MS mass. The anomalous dimension mag is given by
where m
=m
a combination of cm (s ), (s ) and Ccm () and reads

2
s
mag
mag s
mag s
mag =
+
0 + 1
+ 2

2
(1) 2 s 3

(1) s
(2)
(1)
+ 40 ccm
+ 21 ccm
20 ccm
= cm + 20 ccm
(1) 2
(1) 3

(1)
(2)
+ 22 ccm 21 ccm + 20 ccm + 41 ccm

s 4
(1) (2)
(3)
60 ccm
ccm + 60 ccm
+ .
(49)
The terms containing the -function stem from the running of s .

Since Cmag is still multiplied by the inverse pole mass, 1/mQ , in a further step the renormalization group invariant mass M RGI defined by [37]
M RGI = m
s (m
)
20
m,0
0

)
s (m
exp
0

m m,0 ds
0
s
can be used, where we introduced the quark mass anomalous dimension m

2
s
s
s
m =
+ ,
m,0 + m,1
+ m,2
(50)
(51)
with m,0 = 3CF /4; the other coefficients can be found in Refs. [38,39]. Using Eq. (50) in addition to the MS-on-shell relation the overall and logarithmic dependence on the pole mass can
be replaced by the RGI mass. This procedure turns Cmag into Cspin and one obtains an equation
analog to Eq. (48) for Cspin with the anomalous dimension given by
spin = mag 2m
(52)
with
mag
mag
= 0
mag
1
mag
2
mag
3
mag
= 1
mag
= 2
mag
= 3
20 k1 ,

21 k1 + 0 2k12 40 k1 4k2 ,

(1)
22 k1 + 1 2k12 40 k1 4k2 1202 k1 ccm

+ 0 30 k12 2k13 61 k1 + 6k1 k2 60 k2 6k3 .
(53)
Q and the coefThe terms containing the -function stem from the transformation of mQ to m
ficients ki are defined through mQ /m
Q (m
Q ) = 1 + k1 s / + k2 s2 / 2 + k3 s3 / 3 + . They
can be found in Ref. [40] (see also Refs. [4143,16]).
289
Fig. 2. Cmag and Cspin as functions of QCD /mQ for nl = 0 (left) and nl = 4 (right). The upper group of lines corresponds to Cspin while the lower shows Cmag . Inside these groups the dotted, dashed and solid lines show the LO, NLO
and NNLO, respectively. The estimation for the NNNLO result is shown by the dash-dotted line. The region relevant for
the bottom quark is indicated by the vertical line.
In Fig. 2 the results for Cmag and Cspin are shown as a function of QCD /M RGI for nl = 0
(left) and nl = 4 (right) where the LO, NLO and NNLO results are shown. We also added
an estimation for the NNNLO result by assuming a vanishing four-loop anomalous dimension
which is motivated by the smallness of the higher order terms in Eq. (40). For the abscissa we
choose QCD = 0.238 which results from a lattice calculation for nl = 0 [44] and vary the value
for M RGI .
For QCD /M RGI 0.03, which is the range relevant for the bottom quark, one observes in the
case of Cmag a relatively big shift when going from LO to NLO. However, the additional shifts
after including the NNLO and the (estimated) NNNLO are smaller. The convergence improves
significantly when going from Cmag to Cspin . In the case of the bottom quark the NLO corrections
turn out to give a tiny contribution for nl = 0, however, also the NNLO and NNNLO results are
quite small. Going to smaller quark masses one observes moderate corrections for Cspin whereas
for Cmag one has no convergence. We would like to mention that in the case nl = 4 there is
basically no change in the behaviour of Cmag . However, as far as Cspin is concerned one observes
a moderate shift when including the NLO terms whereas the NNLO and NNNLO corrections are
tiny.
6. Magnetic moment
The calculation of the anomalous magnetic moment of a heavy quark proceeds along the
same lines as for the chromomagnetic moment. The only difference is that the external gluon is
replaced by a photon. Our result reads
(n )
aQ
s f (mQ )
=
CF
QQ
2
(nf )

s (mQ ) 2
3
5 2 1 2
31
+
CF + ln 2 + 3 CF
16 12
2
4

317 1 2 1 2
3
25
119 1 2
+
+ ln 2 3 CA + nl +
TF
144 8
4
8
36
36
3
(nf )
3

s (mQ )
(3)
+
aQ + O s4 .
(54)
290
QQ is the charge of the heavy quark in terms of the positron charge. Note that the strong coupling
in Eq. (54) is defined for nf active flavours and it is evaluated at the scale = mQ . The twoloop contribution was already computed in Ref. [45]. Recently, it has also been obtained by
considering the on-shell limit in the calculation of the off-shell form factor [14]. We are in full
agreement with Ref. [14] while we disagree with Ref. [45] by an overall factor of four in the
coefficient of (s /)2 .
Our new three-loop term is given by

20
5 4
5
22
235
103 2
(3)
aQ
=
a4 +
ln 2 2 ln2 2 2 ln 2
5 +
3
3
18
18
3
24
72

139 4 241
23 2 101
+
3 +
CF3
2160
24
6
64

20
5 4
49
31
185
5
a4 +
ln 2 2 ln2 2 + 2 ln 2
5 + 2 3
3
18
36
12
24
12

35 4 113
1505 2 955
+
+
3
+
CF2 CA
432
48
432
72

5
5 4
11
25
65
29 2
+
a4 +
ln 2 2 ln2 2 + 2 ln 2 5 +
3
3
72
18
8
32
288

103 4 1
463 2 31231
+
3
+
CF CA2
2880
2
216
2592

8
1
2
5
11 4
a4 + ln4 2 + 2 ln2 2 2 ln 2
+ 33
3
9
9
3
216

79
125
+ 2
CF2 TF nl
54
32

4
1 4
1
5
11 4 19
+
a4 +
ln 2 + 2 ln2 2 2 ln 2
+ 3
3
18
9
6
432
12

77 2 2411
+

CF CA TF nl
216
324

32
4
4
16
4 4 263
+
a4 + ln4 2 2 ln2 2 2 ln 2 +

3
3
9
9
9
135
72

11 2 7703
+
+
CF2 TF
162
864

20
5 4
5
32
25
1
a4 +
ln 2 2 ln2 2 2 ln 2 + 5 2 3
3
18
18
9
24
8

143 4 241
1375 2 2117
+
+
3 +

CF CA TF
2160
36
648
162

1 2
317
61
+
(nl + 1) +
nl
CF TF2 nl
27
324
162

d abc d abc
8
4 2 943
+
(55)
3

CF TF2 + F F Xsin ,
3
135
324
NF
291
where dFabc is the symmetrized trace of three generators in the fundamental representation (for
SU(Nc ), dFabc dFabc = (Nc2 1)(Nc2 4)/(16Nc )). Xsin denotes the contribution from singlet diagrams (cf. Fig. 1(g)). It is given by
2 4
2
5
5
41 4
ln 2 2 ln2 2 24 2 ln 2 + 5 2 3
3
3
6
18
540
4
931 2 5
3 +
(56)
+ ,
3
54
9
where we only include the contribution from diagrams with closed heavy-quark loops. In principle there are contributions from diagrams with massless quarks as well. However, within
perturbation theory they are divergent. Their evaluation either requires non-perturbative methods
or phenomenological models describing the interaction of light mesons in intermediate states.5
This is in analogy to the light-by-light contribution to the anomalous magnetic moment of the
muon [46], which contains logarithms of the electron mass.
It is possible to obtain the known three-loop result for the electron anomalous magnetic
moment from the expressions in Eqs. (54)(56) by setting CF = TF = 1, CA = 0, NF = 1,
dFabc dFabc = 1 and nl = 0. This result was first obtained in analytical form in Ref. [20] and confirmed in Ref. [15].
Let us evaluate the quark magnetic moment numerically for charm, bottom and top quarks.
Inserting the numerical values for the coefficients, we find
(n )
2
aQ
(n )
= 0.2122s f (mQ ) + (0.8417 0.0469nl ) s f (mQ )
QQ

(n )

3

+ 4.5763 0.5856nl + 0.0145n2l s f (mQ ) + O s4 .
(57)
Xsin = 16a4 +
Using s(4) (mc ) = 0.3378, s(5) (mb ) = 0.2169 and s(6) (mt ) = 0.1075 we obtain
ac = 0.0478 + 0.0533 + 0.0758 = 0.1770,
ab = 0.0153 0.0103 0.0084 = 0.0340,
at = 0.0152 + 0.0047 + 0.0017 = 0.0215.
(58)
The pattern is very similar to the chromomagnetic moment of the heavy quark: no convergence
is observed for the charm quark, the corrections for the bottom quark are large and amount to
more than 50% of the one-loop contribution, and in the top quark case, we find that our new term
gives a 10% contribution which provides quite some confidence that the uncertainty of the final
prediction for at is small.
As already mentioned in the Introduction the LEP1 bound of Ref. [13] for the bottom quark
ab /Qb < 0.045 (68% C.L.)was found to be saturated by the two-loop correction. It is therefore
interesting to see what happens if we include our three-loop term. For this purpose, we have to
evaluate ab for = mZ . We find
ab (mZ ) = 0.0083 0.0066 0.0056 = 0.0206.
(59)
Since the three-loop correction is almost as large as the two-loop one this overshoots the bound
by about 25%. In this context we want to mention again that the contributions from closed lightquark loops could not be included in our calculation. These contributions might decrease the
5 In principle it is possible to calculate the contributions from massive charm and bottom quarks to a and a , respect
b
tively. However, since the integrals involved contain two mass scales, this is beyond the scope of this work.
292
three-loop correction. In any case, a more precise measurement of ab would certainly be interesting.
7. Conclusion
Our main results are the three-loop anomalous dimension of the HQET chromomagneticinteraction operator (13) and the three-loop coefficient of this operator in the HQET Lagrangian (14). They can be used for the investigation of various effects of the heavy-quark spin
symmetry violation (e.g., mass splittings) using continuum or lattice techniques. Furthermore,
we have obtained the three-loop anomalous magnetic moments of heavy quarks (54). This result
does not include the light-quark light-by-light contribution which cannot be calculated perturbatively.
Acknowledgements
We would like to thank Rainer Sommer for fruitful discussions, many explanations in connection to Cmag and Cspin and carefully reading the manuscript. We also thank Luminita Mihaila for
discussions about the colour factors. The work of J.P. was partially supported by NSERC. This
work was supported by the DFG through SFB/TR 9. The Feynman diagrams were drawn with
JaxoDraw [47].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
E. Eichten, B.R. Hill, Phys. Lett. B 243 (1990) 427.

A.F. Falk, B. Grinstein, M.E. Luke, Nucl. Phys. B 357 (1991) 185.
A.V. Manohar, M.B. Wise, Heavy quark physics, Cambridge Monogr. Part. Phys. Nucl. Phys. Cosmol. 10 (2000) 1.
A.G. Grozin, Heavy quark effective theory, Springer Tracts Mod. Phys. 201 (2004) 1.
M.E. Luke, A.V. Manohar, Phys. Lett. B 286 (1992) 348, hep-ph/9205228.
L.F. Abbott, Nucl. Phys. B 185 (1981) 189.
G. Amors, M. Beneke, M. Neubert, Phys. Lett. B 401 (1997) 81, hep-ph/9701375.
A. Czarnecki, A.G. Grozin, Phys. Lett. B 405 (1997) 142, hep-ph/9701415.
A.G. Grozin, M. Neubert, Nucl. Phys. B 508 (1997) 311, hep-ph/9707318.
A.I. Davydychev, A.G. Grozin, Phys. Rev. D 59 (1999) 054023, hep-ph/9809589.
R. Sommer, hep-lat/0611020.
D. Guazzini, H.B. Meyer, R. Sommer, ALPHA Collaboration, arXiv: 0705.1809 [hep-lat].
R. Escribano, E. Mass, Nucl. Phys. B 429 (1994) 19, hep-ph/9403304.
W. Bernreuther, R. Bonciani, T. Gehrmann, R. Heinesch, T. Leineweber, P. Mastrolia, E. Remiddi, Phys. Rev.
Lett. 95 (2005) 261802, hep-ph/0509341.
K. Melnikov, T. van Ritbergen, Nucl. Phys. B 591 (2000) 515, hep-ph/0005131.
P. Marquard, L. Mihaila, J.H. Piclum, M. Steinhauser, Nucl. Phys. B 773 (2007) 1, hep-ph/0702185.
P. Nogueira, J. Comput. Phys. 105 (1993) 279.
R. Harlander, T. Seidensticker, M. Steinhauser, Phys. Lett. B 426 (1998) 125, hep-ph/9712228.
T. Seidensticker, hep-ph/9905298.
S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417.
S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
P. Marquard, D. Seidel, unpublished.
C. Bauer, A. Frink, R. Kreckel, cs.SC/0004015.
R.H. Lewis, Fermats User Guide, http://www.bway.net/~lewis.
M. Tentyukov, J.A.M. Vermaseren, Comput. Phys. Commun. 176 (2007) 385, cs.SC/0604052.
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
T. van Ritbergen, A.N. Schellekens, J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 41, hep-ph/9802376.
K.G. Chetyrkin, B.A. Kniehl, M. Steinhauser, Nucl. Phys. B 510 (1998) 61, hep-ph/9708255.
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
293
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Comput. Phys. Commun. 133 (2000) 43, hep-ph/0004189.
I.I.Y. Bigi, M.A. Shifman, N.G. Uraltsev, A.I. Vainshtein, Phys. Rev. D 52 (1995) 196, hep-ph/9405410.
A.G. Grozin, hep-ph/0008300.
A.G. Grozin, A.V. Smirnov, V.A. Smirnov, JHEP 0611 (2006) 022, hep-ph/0609280.
A.G. Grozin, T. Huber, D. Maitre, JHEP 0707 (2007) 033, arXiv: 0705.2609 [hep-ph].
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379, hep-ph/9701390.
M. Czakon, Nucl. Phys. B 710 (2005) 485, hep-ph/0411261.
W.M. Yao, et al., Particle Data Group, J. Phys. C 33 (2006) 1, http://pdg.lbl.gov.
J. Heitger, A. Juttner, R. Sommer, J. Wennekers, ALPHA Collaboration, JHEP 0411 (2004) 048, hep-ph/0407227.
K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161, hep-ph/9703278.
J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327, hep-ph/9703284.
N. Gray, D.J. Broadhurst, W. Grafe, K. Schilcher, Z. Phys. C 48 (1990) 673.
K.G. Chetyrkin, M. Steinhauser, Nucl. Phys. B 573 (2000) 617, hep-ph/9911434.
K.G. Chetyrkin, M. Steinhauser, Phys. Rev. Lett. 83 (1999) 4001, hep-ph/9907509.
K. Melnikov, T. van Ritbergen, Phys. Lett. B 482 (2000) 99, hep-ph/9912391.
S. Capitani, M. Luscher, R. Sommer, H. Wittig, ALPHA Collaboration, Nucl. Phys. B 544 (1999) 669, heplat/9810063.
[45] J. Fleischer, O.V. Tarasov, Phys. Lett. B 283 (1992) 129.
[46] S. Laporta, E. Remiddi, Phys. Lett. B 301 (1993) 440.
[47] D. Binosi, L. Theussl, Comput. Phys. Commun. 161 (2004) 76, hep-ph/0309015.
Nonsupersymmetric brane/antibrane configurations

in type IIA and M theory
Joseph Marsano a, , Kyriakos Papadodimas b , Masaki Shigemori a
a California Institute of Technology 452-48, Pasadena, CA 91125, USA
b Institute for Theoretical Physics, University of Amsterdam, Valckenierstraat 65, 1018 XE Amsterdam, The Netherlands
Received 15 June 2007; accepted 20 August 2007

Abstract
We study metastable nonsupersymmetric configurations in type IIA string theory, obtained by suspending
D4-branes and D4-branes between holomorphically curved NS5s, which are related to those of hepth/0610249 by T -duality. When the numbers of branes and antibranes are the same, we are able to obtain
an exact M theory lift which can be used to reliably describe the vacuum configuration as a curved NS5
with dissolved RR flux for gs 1 and as a curved M5 for gs 1. When our weakly coupled description is reliable, it is related by T -duality to the deformed IIB geometry with flux of hep-th/0610249 with
moduli exactly minimizing the potential derived therein using special geometry. Moreover, we can use a
direct analysis of the action to argue that this agreement must also hold for the more general brane/antibrane
configurations of hep-th/0610249. On the other hand, when our strongly coupled description is reliable, the
M5 wraps a nonholomorphic minimal area curve that can exhibit quite different properties, suggesting that
the residual structure remaining after spontaneous breaking of supersymmetry at tree level can be further
broken by the effects of string interactions. Finally, we discuss the boundary condition issues raised in hepth/0608157 for nonsupersymmetric IIA configurations, their implications for our setup, and their realization
on the type IIB side.
E-mail addresses: marsano@theory.caltech.edu (J. Marsano), kpapado@science.uva.nl (K. Papadodimas),

shige@theory.caltech.edu (M. Shigemori).
doi:10.1016/j.nuclphysb.2007.08.010
J. Marsano et al. / Nuclear Physics B 789 (2008) 294361
295
1. Introduction and summary

Following the publication of [1], the past year has seen a great deal of interest in the study
of metastable supersymmetry-breaking vacua in supersymmetric gauge theories [27] and string
theory [832]. Because such configurations do not correspond to true vacua, many of the difficulties associated with the construction of realistic models of supersymmetry-breaking can
be avoided. As such, the ideas of [1] have already found wide phenomenological application
[3341].
One particularly interesting system, proposed in [16], realizes metastability by wrapping
branes and antibranes on vanishing 2-cycles of a CalabiYau threefold. The geometry can be
engineered so that these 2-cycles are homologous but, nevertheless, attain a finite size away from
the singular points, providing a barrier to brane/antibrane annihilation. Unlike previous examples, this setup is inherently stringy in that the decay process cannot be described by a quantum
field theory with a finite number of degrees of freedom.1
A particularly nice feature of this system is that it appears to be under fairly good calculational
control. When the numbers of branes and antibranes are large, the authors of [16] suggest that one
can apply the large N duality story of [4246] even in this nonsupersymmetric setting. In other
words, we can effectively replace the branes by a deformed geometry with fluxes. Moreover, it
is argued that the N = 2 supersymmetry on the deformed CalabiYau is actually spontaneously
broken by the opposite-sign fluxes, implying that both the superpotential and Khler potential
continue to be determined by special geometry. This makes it possible to perform controlled
computations and study, for instance, the stabilization of the compact complex structure moduli.
1.1. Brane/antibrane configurations in type IIA
It has been well known for many years that geometrically engineered systems of this sort
are T -dual to HananyWitten type NS5/D4 configurations in type IIA theory [4749]. In this
description, one studies the vacuum configuration by noting that the NS5/D4 system can also
be described by a single M5-brane extended along a potentially complicated 6-dimensional hypersurface. If one tunes the parameters appropriately, the M5 worldvolume theory is reliably
approximated by the NambuGoto action and hence the vacuum configuration corresponds to an
M5 extended along a minimal area surface.
What can one hope to gain from such a description? In supersymmetric examples, BPS arguments indicate that the (holomorphic) physics should not depend on gs so one expects that the
vacuum configuration at all values of the coupling is simply that obtained by applying T -duality
to the IIB picture. We can verify this directly from the M-theory point of view by noting that
the minimal area M5 reduces, at small gs , to a curved NS5-brane with flux which is T -dual to
the IIB deformed geometry in the supersymmetric vacuum [50,51]. This provides a nice alternative way of understanding the large N duality story of type IIB but does not teach us anything
fundamentally new about the physics.
1 If one attempts to decouple stringy modes to get a gauge theory description, one needs to take 0, which in turn
means that one has to simultaneously scale the distance between branes and antibranes to infinity in order to render the
open string tachyon massive. Conversely, keeping the branes and antibranes a finite distance apart, as was considered in
[16], implies that must also be finite and, indeed, is sufficiently large that a field theory description is available only
for the deep IR physics near either the brane or the antibrane stack.
296
(a)
(b)
Fig. 1. The NS5/D4/D4 configuration under consideration. (a) NS5/D4/D4 configuration T -dual to the brane/antibrane
system of [16]. (b) Tilting of D4s and D4s that gives rise to logarithmic bending along the w-direction.
In the nonsupersymmetric case at hand, though, we do not know a priori whether the vacuum
configuration is protected as one moves to different parameter regimes or not. This has two important consequences. First, it means that we must take care to understand the specific choices of
parameters for which a description based on minimal area M5s is valid. Second, it indicates that
the physics in one regime, say strong coupling, need not resemble that in another, say weak coupling, and hence there may be something new to be learned from a IIA/M description depending
on precisely when we can perform reliable computations there.
In this paper, we thus endeavor to use techniques of M theory to study the nonsupersymmetric brane/antibrane configurations that are obtained by applying T -duality to the system of [16],
namely that with D5s and D5s placed at conifold singularities in a local CalabiYau in type
IIB.2,3 Specifically, the setup on which we focus most of our attention is that illustrated in
Fig. 1(a) and consists of a pair of quadratically curved NS5-branes with stacks of D4s and D4s
suspended between them. From the point of view of M theory, this system is described by a single M5-brane which, for parameter regimes in which the NambuGoto piece of the worldvolume
theory is reliable, simply wraps a minimal area surface. There are two distinct parameter regimes
for which this is the case and an analysis based on minimal area surfaces is justified. One lies
at strong coupling, where the x 10 radius is large and the M5 curvature small in 11-dimensional
Planck units. The other, which does not seem to receive as much attention in the literature,4 lies
at weak coupling. There, the M5 is more appropriately viewed as an NS5 with dissolved RR flux.
The worldvolume action of the NS5 is simply the dimensional reduction of that of the M5 and
the reduced NambuGoto term is reliable provided the NS5 is weakly curved and a few other
conditions, about which we will have more to say in Section 3.2.2 and Appendix A, are met.
To summarize, once we find a minimal area M5 curve with the right properties, we can use it
to reliably describe our system as a curved M5 at strong coupling or a curved NS5 with flux at
weak coupling provided we make appropriate choices of parameters.
When the numbers of branes and antibranes are equal, we are able to find an exact solution
to the minimal area equations which has a number of interesting properties. First and foremost,
2 Some comments on the IIA description have been made previously in [17].
3 Nonsupersymmetric brane configurations in IIA have been studied before, for example by the authors of [52,53].
Unlike ours, the setups studied there are stable.

4 The reason perhaps is that one typically is interested in the gauge theory limit where the scales of the system become
substringy and the NambuGoto action ceases to be meaningful.
297
if we consider the regime in which the solution reliably describes our system at weak coupling
as a curved NS5 with flux, it simplifies significantly to a configuration that is indeed T -dual to
a deformed CalabiYau geometry with flux in type IIB. Moreover, the moduli of the Calabi
Yau as determined by the minimal area condition in IIA exactly solve the equations of motion
which follow from the IIB potential derived in [16] using large N duality and special geometry.
Consequently, we are able to understand large N duality, even in this nonsupersymmetric context,
from the IIA point of view as the replacement of the configuration of Fig. 1(a) by a curved NS5brane with flux.
Furthermore, we can extend this agreement to more general situations by studying the NS5
worldvolume action directly. In the regime at weak coupling where our analysis is reliable, we
are able to explicitly demonstrate for arbitrary numbers of branes and antibranes that solving the
equations of motion of this system is mathematically equivalent to starting with the deformed
CalabiYau of type IIB and minimizing the potential obtained from special geometry.
That we find such agreement between the IIB and IIA pictures is slightly nontrivial and, for
reasons that we now explain, further supports the idea that these brane/antibrane setups exhibit
a degree of protection, at least at small string coupling. In particular, our IIA analysis implicitly
assumes that the circle on which we perform T -duality is large in string units while reliability
of the computations of [16] requires instead that the dual circle on the IIB side be large. Consequently, the two descriptions we are comparing correspond to quite different parameter regimes.
For supersymmetric situations, one does not worry about this so much because the usual BPS
arguments suggest that the system is protected as one varies this radius. When supersymmetry
is broken, though, this is not expected to be the case unless there is some additional structure
present. As alluded to before, the authors of [16] have argued that the brane/antibrane systems
under consideration still maintain some residual structure from supersymmetry because, at least
at string tree level, it is broken spontaneously via FI terms. It is this fact that must be responsible
for our ability to successfully relate the IIB and IIA stories at weak coupling.
Given this success, it is natural to ask whether or not our system remains protected, in any
sense, as we move to the strong coupling regime. From our exact solution for the lift of Fig. 1(a)
with equal numbers of D4s and D4s, though, it is easy to see that this does not seem to be the
case in general. The reason for this is that our solution exhibits new nonholomorphic features
which are small when the weak coupling interpretation is reliable but which can become large
when the strong coupling interpretation is reliable. The most obvious of these can be understood
by noting that it is favorable for the D4 and D4 stacks to tilt slightly as depicted in Fig. 1(b). This
is because the energy cost associated with increasing their length is balanced by the decrease in
energy achieved when the branes and antibranes move closer together. Such tilting significantly
impacts the entire geometry of the resulting M5 curve because the D4s and D4s pull on and
dimple the NS5s in a nonholomorphic way [54]. In supersymmetric setups, the direction of
this dimpling is transverse to the NS5s and gets combined with the RR gauge potential, or
x 10 coordinate in the M theory language, to form a holomorphic quantity. In the case at hand,
though, the dimpling is no longer transverse to the NS5s and consequently nonholomorphicity
is introduced throughout the curve, even at infinity.5,6
5 This is not the only nonholomorphic deformation of the geometry that arises but it is the simplest to see without
discussing any details of the solution.
6 As we shall explicitly demonstrate, the tilting described here becomes parametrically small when the minimal area
surface reliably describes the system as an NS5 in IIA. When the minimal area surface reliably describes the system as
an M5 in M-theory, this need not be the case as we can choose it to be small or large.
298
Having an exact solution to the minimal area equations in hand permits us to not only see
nonholomorphic features explicitly but also to demonstrate that they are controlled, at least when
our description can be trusted, by two parameters involving gs N and various characteristic length
scales of the geometry. It is important to note that, unlike at weak coupling, we can take these
parameters to be large at strong coupling while maintaining reliability of our description. Hence,
these features are truly present in at least some part of the parameter space and are not simply
an artifact of our formalism breaking down. However, the nice structure of our solution7 seems
to suggest that one can go further and conjecture that the parameters we find are the only ones
relevant for determining the severity of supersymmetry breaking, meaning that the system
remains protected whenever both are small.
To summarize, we find that using the intuition of M theory to view the configuration of
Fig. 1(a) as a single object, namely an NS5 with flux at weak coupling or an M5 at strong
coupling, allows us to not only obtain an alternative understanding of the large N duality of
[16] but also to probe the brane/antibrane system at strong coupling. From this we learn that
various features of this system seem to be protected as the radius of the T -dual circle is varied
but that this protection does not persist throughout the full parameter space. In particular, there
exists a regime at strong coupling where our description is reliable and new nonholomorphic
features become important. In retrospect, perhaps it is not surprising that stringy interactions can
remove the residual structure of the supersymmetry that is spontaneously broken at tree level. It
is gratifying to see this explicitly, though, and to learn something about the physics of metastable
nonsupersymmetric configurations in string theory at strong coupling.
1.2. Metastability of our configurations
Finally, because this system admits IIA and IIB descriptions that we understand well, it provides a nice example in which to study the subtleties pointed out in [12] and how they arise in
the geometric engineering context. A main point of emphasis in [12] is that nonsupersymmetric
configurations engineered in type IIA from NS5s and D4s of the type we consider here have
different boundary conditions, once quantum effects are taken into account, from the supersymmetric configurations into which they can decay. We can see this quite easily by studying the
configuration of Fig. 1(a), taking the numbers of branes and antibranes to be equal for simplicity.
In the nonsupersymmetric configuration, D4s and D4s pull on the NS5s and dimple them as
mentioned above in a manner that extends out toward infinity. The supersymmetric configuration
that remains after the branes annihilate, though, has no such bending because there are no longer
any D4s or D4s to cause it. From this, it is clear that the configurations have dramatically different boundary conditions at infinity and hence should be viewed as states in different quantum
theories.
Does this mean that the nonsupersymmetric configurations we consider are quantum mechanically stable? It appears to us that the answer is no. By annihilating the fluxes on the curved
NS5-brane, the system can indeed lower its energy and consequently it is favorable to do so via
7 In particular, the solution factorizes into a holomorphic piece, roughly coming from the D4s, and an antiholomorphic
piece, roughly coming from the D4s, when both parameters are small. Because each piece is separately holomorphic
with respect to a different complex structure, each is individually supersymmetric but with respect to different sets of
supercharges. In the absence of further backreaction which breaks this factorization, the system thus seems to exhibit
spontaneous breaking of supersymmetry. Of course, our solution is not reliable everywhere so strictly speaking we only
know for sure that this nice structure is exhibited in the parameter regimes discussed in Appendix A.
299
a tunneling process. Once the fluxes are gone, though, one can no longer support the nontrivial
curvature of the NS5-brane and it begins to straighten. This is much like following the quasikink solution of [12] as the kink moves toward infinity.8 Because the kink never actually reaches
infinity in finite time, the system exhibits a runaway behavior. The decay wants to end, but the
final state at which it can end has moved off to infinity and hence left the theory entirely. This
sort of picture has also been recently advocated in [23].
As a result, our configurations are not in the spirit of [1] in that they are not metastable supersymmetry breaking configurations in a supersymmetric theory. They are indeed metastable
but, because of the boundary conditions, the theory in which they live is not supersymmetric. Instead, supersymmetry is broken by a runaway potential in a manner that seems to be qualitatively
similar to [55].
What does all of this mean from the type IIB point of view? There, the bending of NS5s
corresponds to turning on nontrivial NS 2-form B NS while energy stored in the NS5 tension is
identified with the energy of NS 3-form flux H NS . When the RR-fluxes and anti-RR-fluxes
annihilate one another, the nontrivial H NS can no longer be supported and relaxes just as the
NS5s did on the IIA side. The picture of the decay process we had before thus carries over
entirely, complete with runaway behavior.
There is a distinct difference, however, between the philosophy behind typical NS5/D4 constructions in the literature and studies of the local CalabiYau configurations to which they
are related by T -duality which affects how one interprets these results. When one engineers
gauge theories and other perhaps nonsupersymmetric setups using extended NS5s and D4s in
type IIA, it is usually assumed from the outset that the full theory under consideration is truly
10-dimensional type IIA with fully noncompact branes. This means, for instance, that gauge theories realized in this manner are provided with a specific UV completion from the outset, namely
MQCD.9 That is not to say that a local interpretation of these configurations is not possible but
rather that it does not seem as natural in the IIA context.
On the other hand, the philosophy behind local type IIB constructions is quite different. One
imagines that the local CalabiYau is capturing the physics in a particular region of a larger,
compact CalabiYau. The situation is quite similar to effective field theory in that one imposes
a cutoff scale in order to perform computations and specifies the values of noncompact moduli,
which play the role of coupling constants, at that scale. Of course, one can take the full noncompact CalabiYau seriously by taking the cutoff scale to infinity.10 This would correspond to
UV-completing the effective local description into the T -dual of MQCD. Typically, though, this
is not our interest as compact situations are more realistic for practical applications.
Nevertheless, we learn something very important about these local IIB constructions from the
subtleties that arose in type IIA. While we can study the tunneling process by which the fluxes
and anti-fluxes annihilate in a local context, the eventual endpoint of the decay is highly dependent on how we UV-complete the local configuration into a compact CalabiYau. This is already
evident from the computations in [16], where the energy difference between the supersymmetric
and nonsupersymmetric configurations computed in a regularization scheme with finite cutoff
exhibits an explicit dependence on that cutoff which cannot be removed. The lesson here seems
8 More precisely, the quasikink of [12] was actually the opposite of what we discuss here with a supersymmetrybreaking configuration in the interior glued to supersymmetric boundary conditions. The idea is the same, however.
9 In particular, this is precisely the situation studied by the authors of [12].
10 One also tries to keep IR quantities fixed in this limit. Quantities for which this is possible are well described by
effective field theory.
300
to be that, while local constructions are useful for studying some aspects of metastable systems
in string theory, one must be careful of the inherent limitations of such descriptions and take care
to ask questions which they are well suited to answer.
1.3. Outline of the paper
The organization of this paper is as follows. In Section 2, we introduce the type IIB and IIA
constructions that shall be the focal point of our work and review their relation to one another via
T -duality. In Section 3, we review the relation between these approaches in two supersymmetric
examples. In the first, we consider branes at a single conifold singularity in type IIB and their
T -dual description in terms of a pair of NS5s with a single stack of D4s suspended between
them. This will allow us to review the basic philosophy behind the construction of parametric
M5 curves. In the second example, we consider branes at a pair of conifold singularities and
their T -dual description in terms of quadratically bent NS5s with two stacks of D4s suspended
between them. This will allow us to introduce a formalism for constructing genus-1 M5 curves
parametrically which will be required when considering nonsupersymmetric configurations. In
Section 4, we turn to the brane/antibrane system of [16], its T -dual description in type IIA,
and the exact M-theory lift. In Section 5, we discuss various things we might learn from our
solution regarding the severity of supersymmetry breaking, address issues related to boundary
conditions and decays in greater detail, and finally mention a few possible future directions.
Appendices AD include various technical details.
2. Preliminaries
In this section, we discuss the local CalabiYau geometries which play an essential role in the
IIB constructions of interest [4345] as well as the type IIA brane setups [54,56] to which they
are related by T -duality [4749]. Both of these are well known, but we review them here for
completeness and in order to make precise the specific T -duality dictionary we shall be using.
2.1. IIB geometric constructions
We begin by considering the A1 ALE space, which can be realized by the complex equation
x 2 + y 2 + w 2 = 0,
x, y, z C.
(2.1)
One can view this as a C fibration over the w plane as in Fig. 2. The singularity at w = 0
corresponds to a singular double-degeneration of the nontrivial S 1 in C and can be removed by
introducing a complex deformation as follows
x 2 + y 2 + w 2 = 2 .
(2.2)
The singular degeneration of S 1 at w = 0 has now been replaced by smooth degenerations at w =

. Fibering the S 1 over this interval yields an S 2 which has grown in place of the singularity as
depicted in Fig. 2. We can now construct a local CalabiYau threefold by fibering this deformed
A1 surface over a plane parametrized by a fourth complex parameter, v C, as in Fig. 3. This is
easily accomplished in (2.2) by replacing the constant with a holomorphic function W (v)
x 2 + y 2 + w 2 = W (v)2 .
(2.3)
301
Fig. 2. Singular and deformed A1 singularities.
Fig. 3. Singular and deformed A1 fibration.
For generic values of v, the S 2 at the tip of the cone has finite volume but it degenerates at the
zeros of W (v). Moreover, because the roots of W (v)2 = 0 necessarily have multiplicity at least
two, this degeneration is singular. To deal with this, one can proceed by analogy to what we did
for A1 itself, namely introduce a complex deformation that breaks the double-degeneracy
x 2 + y 2 + w 2 = W (v)2 f (v).
(2.4)
For generic nontrivial f (v), each singular degeneration point of the S 2 will split into two points
where the degeneration is smooth. Fibering the S 2 over an interval connecting these points then
reveals that a compact S 3 has grown in place of the singularity. We will refer to these 3-cycles
as the A cycles of the geometry. The dual B-cycles of the geometry are noncompact and can be
obtained by fibering the S 2 over an interval beginning at one point of a given pair and extending
to infinity along v. An illustration of this can be found in Fig. 3.
Let us now consider compactifying type IIB on the undeformed local CalabiYau (2.3) and
wrapping D5-branes on the degenerating S 2 . In order to prevent the effective 4d gauge coupling
302
constant on the brane worldvolume from diverging we must turn on a nontrivial NSNS twoform field, B NS , along the shrinking S 2 . A nontrivial -angle can be introduced by turning on
the RR two-form C2RR as well. For a trivial fibration W (v) = 0, the worldvolume theory is simply
N = 2 SYM. A simple expansion of the DBI action reveals that the B and C2 fields determine
the effective 4d complexified coupling via
ib
4i
+ 2 =c+ ,
2
gs
gYM
where
1
b=
4 2
B
S2
NS
fiber
(2.5)
1
c=
4 2
C2RR .
S2
(2.6)
fiber
Note that, as we shall continue to do throughout all that follows, we have set
= 1.
(2.7)
The fields of this theory include an adjoint scalar which parametrizes the locations of the
branes along v. Nontrivially fibering the deformed A1 over the v-plane restricts the branes to sit
at the critical points of W (v), where the S 2 s they wrap degenerate. From the gauge theory point
of view, this nontrivial fibration corresponds to introducing a superpotential W () for the adjoint
superfield11 [43,44]. In what follows, we shall consider superpotentials Wn that are polynomials
of degree n 1.12 In
of degree n + 1 and restrict to deformations fn1 (v) that are polynomials
n
these theories, the gauge group is Higgsed according to U (N ) i=1 U (Ni ) with Ni denoting
the number of branes sitting at the ith critical point of Wn (v).
While this geometric construction provides a nice visualization for the various Higgs branches
that are present in the gauge theory, a direct analysis of the quantum dynamics is not immediately
obvious because it requires going beyond the classical probe approximation for the D5-branes.
It is by now well known that, in order to deal with this, one can use large N duality [43] to
replace D5s at the singular points of the geometry (2.3) with RR 3-form flux H RR wrapping
S 3 s in the deformed geometry (2.4). In addition, we must also introduce some 3-form flux on
the noncompact B-cycles of (2.4) in accordance with the B NS and C2 that threaded the vanishing
S 2 s of (2.3). The degrees of freedom of this system include n chiral superfields associated to the
sizes of the S 3 s and n Abelian vector superfields obtained by reducing the RR 4-form potential.
The former are identified with glueball superfields associated to the confined SU(Ni ) factors
while the latter correspond to the n spectator U (1)s.
Once we have replaced our brane configuration by deformed geometry with fluxes, the quantum dynamics becomes easy to study because it is captured by the well-known GukovVafa
Witten (GVW) superpotential [57]

W = H ,
(2.8)
where is the holomorphic 3-form and H is a combination of the NSNS and RR 3-forms
H = H RR +
i NS
H .
gs
(2.9)
11 As the notation suggests, the function W (v) that appears in the local CalabiYau (2.3) is nothing more than the
derivative of this superpotential [44].
12 In other words, we restrict to normalizable and log-normalizable complex structure deformations [43].
303
Right away, however, we notice that noncompactness of the B-cycles leads to problems because they have infinite holomorphic volume. In other words, the B-periods of , which appear
directly in (2.8), are divergent. In order to make (2.8) meaningful in practice, then, we must impose an arbitrary cutoff v0 on integrals over the base. Concurrently, it is also necessary to specify
the boundary conditions of the system at this cutoff scale. We can accomplish this by fixing the
integrals of the 2-form potential over the S 2 fiber at v0
v0

i NS
i
RR
2
C2 + B
= 4 c(v0 ) + b(v0 ) ,
gs
gs

H=
Bi
S2
(2.10)
fiber at v0
which is equivalent to specifying the regulated H -flux along the noncompact B-cycles. It is
possible to incorporate v0 -dependence in the boundary conditions c(v0 ), b(v0 ) in such a manner
that explicit cutoff-dependence is removed from the superpotential (2.8). From the gauge theory
point of view, this entire procedure is well known. We have simply introduced a UV cutoff scale
and specified the values of our coupling constants at that scale. Changes in cutoff scale must
be accompanied by shifts in the couplings consistent with RG flow if we wish to preserve the IR
physics.
Finally, we note that it is often useful in practice to integrate H and over the S 2 fiber in
order to reduce the problem to one involving 1-forms defined on a Riemann surface. We can do
this explicitly for 13
1
=
2

=

dv
Wn (v)2 fn1 (v).
2
(2.11)
S2
This 1-form is well defined on the hyperelliptic curve

w 2 = Wn (v)2 fn1 (v),
(2.12)
which can be visualized as a double-cover of the v-plane with cuts connecting the various S 2
degeneration points. The A and B cycles of the local CalabiYau (3.1) now descend to A and B
cycles on this curve.14 A convenient basis for these cycles is depicted in Fig. 4.
On the other hand, we do not have an explicit expression for H or its corresponding reduced
1-form15
1
h
2

H
(2.13)
S2
13 We insert the factor of 1 for convenience in order to absorb the factor of 2 difference between the A and B cycles of
2
the local CalabiYau, which pass along each cut once, and the A and

of the

hyperelliptic curve, which
B cycles
reduced
encircle each cut, effectively passing along it twice (in equations, A = 12 A S 2 , B = 12 B S 2 ).
14 This is true up to the usual factors of 2, which we have absorbed into the definitions of h and .
15 The minus sign is inserted for convenience only.
304
Fig. 4. Sample hyperelliptic curve for n = 3 with A and B cycles indicated.
but knowledge of the fluxes determines its periods along nontrivial cycles of (2.12) as follows16,17

1
2 i
H = 4 N
h = 2iN i ,
(2.14)
2i
Ai
v0
Ai
H = 4 2 (v0 )
Bi
1
2i

h = 2i(v0 ),
(2.15)
Bi

(v0 ) c(v0 ) +
ib(v0 )
4i
= 2 +
.
gs
gYM 2
(2.16)
2.2. IIA brane constructions

Gauge theories that can be engineered using the type IIB constructions reviewed in the previous section can also be obtained from brane configurations in type IIA involving NS5s and D4s
[54,56,58], as we now briefly review. In the following, we shall consider configurations with two
NS5-branes and N D4-branes extended along the 0123 directions. The NS5s also extend along
holomorphic curves of the form w = w(v) where
v = x 4 + ix 5 ,
w = x 7 + ix 8
(2.17)
and are separated along x 6 by a distance L. The D4s are then suspended between the NS5s
along x 6 .
Let us begin by considering the case of parallel NS5-branes wrapping the curves w = 0 in
wv space. This is the situation depicted in Fig. 5. If we scale the length L of the D4-branes
along x 6 to zero, the theory on their worldvolume becomes effectively four-dimensional with
16 The factor of 4 2 here is simply the fundamental unit of 3-form flux, 2 2 , sourced by a single D5 brane.
10 5
17 Note that we use upper indices N i for the fluxes as opposed to the lower indices N denoting the number of branes.
i
For branes, these are the same but for antibranes the sign is opposite.
305
Fig. 5. A sample NS5/D4-brane configuration which realizes N = 2 SYM on the D4 worldvolume.
gauge coupling constant given by

8 2
L
= .
2
gYM
gs
(2.18)
The brane configuration preserves 8 supercharges so this theory has N = 2 supersymmetry in

four dimensions. The N = 2 vector superfield consists of an N = 1 vector superfield as well as
an adjoint chiral superfield which parametrizes the location of the D4s along v.
We can now introduce a superpotential W () for the adjoint superfield by extending the
NS5-branes instead along the nontrivial curves [56,5860]
w(v) = W (v).
(2.19)
The various ways of distributing D4-branes at the critical points of W (v) correspond to the
different Higgs branches of the theory.
While the classical brane picture is useful for visualizing the Higgs structure it is easy to see
that, like the wrapped D5-brane configuration of the previous section, it obscures the quantum
dynamics. The reason for this is that the setup does not describe a key element of the quantum
system, namely backreaction of the D4s on the NS5s. In particular, it is well known that the
NS5 throat is a region of large string coupling so it is
difficult to analyze the NS5/D4 intersection,
which plays a dominant role particularly when L , in a IIA context.
To deal with this, Witten noted that NS5s and D4s are two different manifestations of the
same object, namely the M5-brane, and hence their intersection could be smoothed out by looking at this system from the point of view of M theory [54]. There, our NS5/D4 configuration is
thought of instead as a single M5-brane extended along a possibly complicated 6-dimensional
hypersurface. At large gs and small 11-dimensional Planck length, 11 , the worldvolume theory
of the M5 is effectively described by the NambuGoto action so its embedding into target space
is one of minimal area. At small gs , on the other hand, this M5 is better thought of as a curved
NS5-brane with dissolved RR flux whose worldvolume theory is obtained by dimensional reduction. When the descendant of the NambuGoto term gives a reliable description of physics in
the IR, this NS5 configuration can be obtained directly from the M5 one at large gs by reducing
along the M-circle.18 In practice, this simply means that we reinterpret the coordinate x 10 in our
M5 solutions as the appropriate RR gauge potential.
18 For supersymmetric configurations, one typically performs this IIA reduction without a second thought as the M5
is protected in such cases from corrections that arise as gs is decreased. Because we are eventually interested in nonsupersymmetric configurations, we shall always be careful to specify when using the NambuGoto term at small gs is
reliable.
306
In the situation at hand, the M5-brane extends along the 0123 directions and, due to supersymmetry, wraps a holomorphic curve in the remaining directions. Because each D4 stack in
the IIA configuration fattens into a tube upon lifting to M theory, the genus of is related to the
number n of such stacks by g = n 1. It is convenient to use one of the complex coordinates,
say v, to parametrize this curve, permitting us to think of it as a double cover of the v-plane with
n cuts.19
To determine the correct lift, we must impose the boundary conditions (2.19) along the w and
v directions as well as require that the wrapping along x 10 be consistent with the D4 distribution
in IIA. Turning first to w(v), a hyperelliptic curve satisfying w(v) W (v) as v can be
written in the form

w(v) = W (v)2 fn1 (v).
(2.20)
Note that this is precisely the curve on which the 1-form (2.11) was defined in the previous
section. As such, we will continue to the basis of A and B cycles indicated in Fig. 4.
To deal with the x 10 constraint, we first combine x 10 with x 6 into a holomorphic variable [54]

s = R 1 x 6 + ix 10 ,
(2.21)
where R = gs is the radius of the M-circle. This is crucial because in the end we are looking
for a holomorphic curve. With this definition, the condition that x 10 wrappings be consistent with
the distribution of D4s in the IIA reduction can be expressed as a constraint on the A-periods of
the 1-form ds

(2.22)
ds = 2iNi .
Ai
The B-periods of ds, on the other hand, are related to the separation between the NS5s along x 6
(and potentially along x 10 as well). Generically, x 6 will vary logarithmically with v due to the
fact that the D4s pull on and dimple the NS5 [54]. As a result, integrals of ds along the
noncompact B-cycle will in general diverge, forcing us to introduce a cutoff on the v-integration
at an arbitrary point v0 . This corresponds to introducing a UV cutoff in the gauge theory on the
D4 worldvolume. We identify this separation with the (complexified) 4d gauge coupling at this
scale
v0
8 2
(2.23)
ds = 2i(v0 ) = 2
+ ii (v0 )
gYM (v0 )
Bi
and hence the dependence on v0 simply corresponds to RG flow. Once we have determined the
holomorphic curve , we can reduce to IIA by simply reinterpreting x 10 , thus obtaining the
curved NS5-brane with flux that results from backreaction of the D4s.
The M-theory description, as we have reviewed it so far, is purely on-shell so that, unlike the
case of IIB with fluxes, we have no analog of the GVW superpotential that allows one to have
an off-shell understanding of the system. Later, we will see one sense in which this description
can be extended to capture some off-shell information. Another way this can be accomplished
was suggested by Witten, who conjectured a form for the superpotential which captures offshell physics of an M5 wrapping [56]. In particular, if we let 0 denote a reference surface
19 Of course, this relies on the fact that is actually hyperelliptic [54,56].
307
homologous to and B a 3-chain interpolating between the two, the superpotential he wrote is

1
.
W () W (0 ) =
(2.24)
2i
B
Later, de Boer and de Haro [61] noted that this can be rewritten in the form

W ds w dv
(2.25)
which is quite suggestive given the similarities between the 1-forms w dv and ds here and the 1forms (2.11) and h (2.13) of the previous section. Of course, as we now review, this similarity
is not an accident.
2.3. T -duality between IIA and IIB constructions
The above realizations of gauge theories in type IIA and IIB string theories are related by
T -duality [4751]. Here we briefly review this relation and establish the mapping between quantities on both sides.
As we have seen, the type IIB construction is based on fibration of deformed A1 ALE space
(2.2) over the complex v-plane. Consequently, if we understand the T -duality between the ALE
fiber over a point v and the type IIA brane configuration at v, the T -duality relation between the
whole systems will follow by applying it fiberwise. Therefore, we focus here on the T -duality
between ALE space and NS5-branes.
The deformed A1 ALE space whose complex structure is displayed in (2.2) can be realized as
a two-center TaubNUT space (see, e.g., [54, Section 3]).20 The metric of the Euclidean k-center
TaubNUT space is
2
2
dsIIB
= H 1 (d y + )2 + H dz2 + ds
,
H (z) = 1 +
Hp ,
Hp (z) =
p=1
e2IIB = 1,
RIIB
,
2|z zp |
0 y 2RIIB ,
d = 3 dH.
(2.26)
Here zp , p = 1, . . . , k, is the position of the pth center in the base K = R3 parametrized by z =

2 is the metric for the remaining
(x 7 , x 8 , x 9 ) = (Re w, Im w, x 9 ), (z) is a 1-form in K, and ds
six directions 012345. If we T -dualize the TaubNUT metric (2.26) along y using the standard
Buscher rule [62], we obtain the IIA metric

2
2
= H dy 2 + dz2 + ds
,
e2IIA = H,
BIIA = d y.
dsIIA
(2.27)
Here y is the T -dual of y whose periodicity is
y
= y + 2RIIA ,
RIIA =

,
RIIB
(2.28)
and corresponds to x 6 in the last subsection. The metric (2.27) is nothing but the geometry
produced by k NS5-branes located at z = zp in flat space. In particular, if we set k = 2 and
z1,2 = (Re , Im , 0) (i.e., w = ), this shows that the deformed A1 ALE space (2.2) is
20 Strictly speaking, a TaubNUT space is ALF and we must take R
IIB to make it ALE.
308
T -dual to two NS5-branes at w = . Fibering this T -duality over the v-plane, we see that the
local CY space (2.3) is T -dual to two NS5-branes placed along the wv curve (2.19) in a flat
space.
In the metric (2.27), though, the NS5-branes are delocalized (smeared) in the y = x 6 direction.
However, in string theory, the NS5-branes are expected to become localized; indeed, it is known
that the y = x 6 position of the IIA NS5-brane is dual to B-field through certain 2-cycles in the IIB
TaubNUT geometry [63,64]. Although one could study this localization of NS5-branes using
worldsheet CFT techniques [65,66], in Appendix B we have presented an alternative approach to
determining the position of NS5-branes which, to our knowledge, is new.
From (B.12), the 2-form fields in IIB are related to the distance between two NS5-branes in
IIA in the following manner:

i
i
(2.29)
C2RR + IIB B2NS = 4 2 c + IIB b = 4 2 (v0 ) = 2is,
gs
gs
where was defined in (2.16) and s in (2.21). The gauge theory couplings derived in IIB and IIA
(Eqs. (2.5) and (2.18)) can be shown to be identical using this relation, as they should be. From
(2.29) immediately follows also the correspondence between the following 1-forms in IIB and
IIA:
h
ds.
(2.30)
2i
From this relation, it is clear that the periods of h in IIB, (2.14) and (2.15), are mapped into the
periods of ds in IIA, (2.22) and (2.23). If we further note the equivalence between the following
1-forms in IIB and IIA (Eqs. (2.11) and (2.20)):

1
1

w dv,
=
(2.31)
2
2
S2
we can see that the IIB superpotential (2.8) is equivalent to the IIA superpotential (2.25):

WIIB H h WIIA ds w dv.
(2.32)
Summarizing the discussion so far, local CY geometries in type IIB are T -dual to NS5-brane
configurations in IIA, and the realization of gauge theories based on them are equivalent. There
is one important issue that we have glossed over, though. The y and y circles are compact with
radii RIIA and RIIB , respectively, and are related to each other by (2.28). In the IIB and IIA/M
constructions in the previous sections, we treated these circles as if they were noncompact, by
putting D5s in a noncompact (local) CY in IIB and putting NS5s and D4s in noncompact R6
1 ). However, the validity of such a noncompact description is
in IIA (or an M5 in R6 S10
not obvious, because if RIIA is large then RIIB is small, and vice versa. In the supersymmetric
case, if we take the gauge theory limit (decoupling limit) where the scales of the system become
the circle direction
substringy,21 such a noncompact description can indeed be justified because
becomes much larger than the system size if we take RIIA RIIB . What if we do not
take the gauge theory limit? Even then, as long as we focus on holomorphic quantities such
21 In the gauge theory limit in type IIA (IIB), we take the length L of D4 (the size of the S 2 on which D5 is wrapping)
and the distance between different D4 stacks (D5 stacks) to be substringy, such that L gs and E, where E
is the energy scale we are looking at.
309
as the curve (2.12), (2.20), we can still use the noncompact description. This is because these
holomorphic quantities are protected by supersymmetry and do not depend on the scales of the
system such as RIIA , RIIB . Namely, we are free to take them to be infinite. In this sense, the
noncompact IIB and IIA/M constructions in the previous sections are T -dual to each other if
supersymmetry is preserved.
In the nonsupersymmetric case that we shall study later, things can be more subtle. In order
for the fundamental string stretching between D-branes and anti-D-branes to be free of tachyonic
modes,
we must keep the distance between them to be at least of the order of the string length:
of the relation (2.28), it is impossible to make both RIIA and RIIB
. However, because
much larger than at the same time. So, the full physics of the noncompact IIB and IIA/M
constructions is not going to be the same. So, in a strict sense, by studying the noncompact IIA/M
system we will be exploring the nonsupersymmetric physics of a new system which is different
from the IIB system studied in [16,21].
However, even in the nonsupersymmetric case, it is possible that certain quantities are still
protected, if the supersymmetry breaking is soft [67]. For such quantities, the scale parameters
RIIA,IIB are again irrelevant. Therefore, as far as such data are concerned, we can still say that
the noncompact IIB and IIA/M constructions are in fact T -dual to each other and describing the
same physics. Indeed, we will see that certain quantities computed in IIA/M are the same as ones
computed in IIB [16,21], although supersymmetry is broken.
3. Two supersymmetric examples
We now proceed to elaborate upon the connection between the type IIB and IIA constructions
reviewed in the previous section by looking at a pair of simple examples. Among other things,
this will permit us to review the parametric representation of genus zero M5 curves [56] and
introduce the formalism for extending this sort of description to genus one situations in a more
friendly, supersymmetric setting.
3.1. A1 theory with quadratic superpotentialIIB
We begin with perhaps the simplest possible example, namely that of D5-branes at a conifold
singularity [42,43]. The relevant geometry is (2.3) with a quadratic superpotential of the form
W (v) = mv 2 /2. After implementing large N duality, we obtain the deformed geometry
x 2 + y 2 + w 2 = m2 v 2 + f0
(3.1)
with 3-form fluxes on the compact and noncompact cycles

1
1
2
H = 4 N,
H = 4 2 .
2i
2i
A
(3.2)
This geometry has one modulus, f0 , whose value is fixed dynamically. To study this further, we
introduce A and B periods of the holomorphic form as usual

1
F
1
,
=
,
S
(3.3)
2i
2i
S
A
where F is the N = 2 prepotential. The field S serves as an alternate means of parametrizing the
(one-dimensional) moduli space of complex structures and, in fact, its relation to the modulus f0
310
is easy to work out in this simple example

f0
(3.4)
.
4m
The expectation value of S, and hence of f0 , can now be obtained by minimizing the GVW
superpotential (2.8). Using the Riemann bilinear relations, one can rewrite WGVW as

WGVW = H N + S
(3.5)
S=
and immediately obtain the supersymmetric vacuum condition

+ N = 0,
where is the period matrix of the
(3.6)
CalabiYau22 :
2F
(3.7)
.
=
S
S 2
Though the solution = /N to (3.6) provides a perfectly good description of the complex
structure of (3.1) at the supersymmetric vacuum, it is often desirable to translate this into a
statement about the expectation value for S. This requires us to compute (S), a task that is
easily achieved in this simple example. Imposing a cutoff v0 on B-periods as discussed in the
previous section, we find

S
1
ln
+ ,
=
(3.8)
2i
mv02

where terms that vanish as v0 have been dropped. Inverting this and applying (3.6) the
expectation value immediately follows
N

S N = mv02 e2i(v0 ) = mN 2N
(3.9)
n=1 .
In this expression, we have exhibited the explicit v0 -dependence of the coupling constant
that is needed to render WGVW cutoff-independent as well as introduced an RG-invariant scale
n=1 23

8 2
2N
2N 2i(v0 )
2N
+ i(v0 ) .
n=1 v0 e
= v0 exp 2
(3.10)
gYM (v0 )
This completes our brief review of the system obtained by wrapping N D5-branes at a conifold singularity. We have seen that the supersymmetric vacuum is described by the deformed
geometry (3.1) with fluxes (3.2) and complex modulus (3.9).
3.2. A1 theory with quadratic superpotentialIIA/M
We now proceed to study this system from the IIA/M perspective [50,51,56]. Applying T duality, we obtain a brane configuration with two NS5s extended along the curves w = mv
22 We use a rather unconventional notation here, with denoting the period matrix as opposed to . The reason for this
is to avoid confusion later when is used as the complex structure modulus of an auxiliary torus.
23 The subscript n = 1 refers to the fact that, in the theory we are studying, W (v) has degree n = 1. This is to distinguish
n=1 from the analogous quantity introduced later for theories having W (v) of degree 2.
311
Fig. 6. NS5/D4 configuration obtained by applying T -duality to collection of D5-branes at a conifold singularity.
and separated along x 6 with N D4s suspended between. This configuration is depicted in Fig. 6.
As discussed in Section 2.2, in order to describe this system away from gs = 0 we should view
it instead as an M5 extended along a genus zero holomorphic curve with boundary conditions
w mv
as v
and embedding coordinate s, defined in (2.21) to describe the wrapping along

ds = 2iN,
(3.11)
x 10 ,
satisfying
(3.12)
ds = 2i(v0 ).
(3.13)
Though an explicit representation of this curve is well known [56], we shall review the parametric
one here because it will more easily generalize to the nonsupersymmetric curves of interest later.
The curve we seek to study has genus zero so it can be parametrized by a single copy of the
complex plane with a pair of marked points, corresponding to the preimages of on each of the
two NS5-branes. For definiteness, we refer to the complex parameter as and place the marked
points at = 0 and = . At these points, the holomorphic functions w() and v() must
diverge and, moreover, because the embedding is 11 near , these divergences must come in
the form of first order poles. Combined with the boundary conditions (3.11), this is sufficient to
fix their form up to an overall rescaling
a
v() = + ,

a
w() = m
.
From this, it is easy to verify that w and v are related as in (2.20)

w 2 = m2 v 2 4a
(3.14)
(3.15)
and hence that (3.14) provides a parametric description of the hyperelliptic curve depicted in
Fig. 7(a). The A and B cycles shown there appear on the plane as illustrated in Fig. 7(b).
We now turn to the embedding coordinate s, which characterizes wrapping along x 10 . A holomorphic s() with A-period (3.12) is easily seen to be
s = N ln .
(3.16)
312
(a)
(b)
Fig. 7. Parametrizations of the hyperelliptic curve (3.14). (a) Hyperelliptic curve (3.14) with A and B cycles indicated.
(b) plane used to parametrize the hyperelliptic curve (3.14) with A and B cycles indicated.
The logarithmic behavior seen here, which we alluded to in the previous section, necessitates the
introduction of a cutoff v0 in order to study the B-period constraint (3.13)
v0
2i =
2
v
ds = N ln 0 + .
a
(3.17)
From this, we see that a holomorphic 1-form ds with the properties (3.12) and (3.13) exists on
the hyperelliptic curve (3.14) provided the complex parameter a satisfies
a N = v02N e2i(v0 ) = 2N
n=1 ,
(3.18)
where n=1 is as in (3.10). This completely fixes the complex structure of the hyperelliptic
curve (3.14).24
3.2.1. Comparison with IIB
If we interpret (3.14) and (3.16) as describing a curved NS5-brane with flux, the configuration
at hand is T -dual to a deformed geometry of the type (3.1) with flux (3.2). In fact, we can even
verify that the modulus of our IIA configuration is identical to that determined in IIB by dynamics of the GVW superpotential (2.8). One way to do this, for instance, is to check the vacuum
equation (3.6) directly. This can be done because the period matrix of the curve (3.14) is
easy to compute in terms of a

1
a
=
(3.19)
ln 2 + .
2i
v0
Using (3.18), it immediately follows that (3.6) is satisfied. Alternatively, we can make the comparison by computing S directly. The T -duality dictionary provides us with an expression for S
that is easy to evaluate on the IIA side

1
1
S=
(3.20)
w dv = ma = m2n=1 .
2i
2
A
24 Modulo N choices parametrized by N th roots of unity.
313
Using (3.18), we see that the expectation value (3.9) is reproduced. Consequently, we see that
going from the NS5/D4 configuration of Fig. 6 to the curved NS5 with flux described by (3.14)
and (3.16) is exactly T -dual to the large N duality in type IIB [50,51]!
3.2.2. Reliability of the M5 and NS5 descriptions
Now that we have completed our description of the M5 configuration which reduces to that
of Fig. 6 at gs = 0, we must address the question of when this analysis is reliable. What we have
found is a minimal area surface or, in other words, a solution to the equations of motion which
follow from the NambuGoto contribution to the worldvolume action. Viewing our setup as a
curved M5-brane, this is justified provided a number of conditions are met. First, we require that
the curvature of the M5 be everywhere small in 11-dimensional Planck units and the radius of
the x 10 circle large. Furthermore, we must avoid letting N become too large in order to prevent
the density of windings along x 10 from growing to the point that the M5 comes within a Planck
length of intersecting itself. As discussed in Appendix A, one can demonstrate that all of these
conditions are satisfied provided gs 1 and the conditions (A.20) involving N are satisfied.
On the other hand, we can attempt to provide a direct IIA interpretation of our setup as a
curved NS5-brane with flux. Because the NS5 worldvolume action is simply the dimensional
reduction of the M5 one [68], our analysis is valid in the IIA regime, where x 10 is identified with
the appropriate RR gauge potential, provided we can restrict attention only to the descendant of
the NambuGoto term. This, in turn, requires that the curvature of the NS5 be small in string units
and that the flux N not be too large.25 We show in Appendix A that this leads to the constraints
gs 1,

S
n=1
= min
, Sm ,
gs N min n=1 ,
m
m

S
n1
2
3
1 min m n=1 ,
(3.21)
m S,
= min
,
m
m3
where n=1 is as in (3.10) and S is given by (3.20). Physically, these come about because when
m > 1, for example, S/m3 sets the scale of the tubes into which the D4s blow up in the lift
while S/m determines the flux density. The symmetry under m m1 simply reflects our ability
to interchange w v.
Outside of the regimes (A.20) and (3.21), our analysis based on the NambuGoto term of the
worldvolume action is no longer reliable. One still expects the system to be described by an M5
along the curve described by (3.14) and (3.16) but this is dependent on a BPS argument that
relies on supersymmetry.
3.3. A1 theory with cubic superpotentialIIB
We now move on to the second example, namely that of D5-branes at the conifold singularities
of the A1 fibration (2.3) with

3
2 v
v
.
W (v) = g
(3.22)
3
4
25 The reason for this is to prevent excited string states from becoming light. In particular, M2-branes ending on the M5
and wrapping x 10 in the M-theory picture descend in IIA to strings with tension that can be made arbitrarily small unless
gs N is sufficiently small. See Appendix A.
314
In particular, we have two singularities at v = /2 at which we place N1 and N2 D5s, respectively. After the geometric transition, we are left with the deformed geometry [43]

2 2
x 2 + y 2 + w2 = g 2 v 2
(3.23)
f 1 v f0
4
and 3-form fluxes

H = 4 2 N i ,
Ai

H = 4 2 i ,
(3.24)
Bi
where 1 = 2 =
This system has two complex moduli corresponding to the holomorphic
volumes of the compact S 3 s

1
,
Si =
(3.25)
2i
.26
Ai
which can in turn be related to the complex deformation parameters f0 and f1 , though we do not
do it explicitly here. Introducing the B-period i and period matrix ij as usual

1
F
i
2F
i
(3.26)
=
,
ij
=
,
2i
Si
Sj
Si Sj
Bi
we can write the GVW superpotential as

WGVW Ni i + i Si
(3.27)
and obtain the supersymmetric vacuum condition

i + ij Nj = 0.
(3.28)
This specifies ij in terms of the fluxes , Ni and hence completely fixes the complex moduli.
To translate this into a statement about the Si , it is necessary to determine the dependence of
ij on the Si . This can be done using the results of [43], who compute the prepotential F as an
expansion in the variables
ti =
Si
.
g3
(3.29)
Applying their result, we find that ij is given by the following to leading order in the tk

2
1
v0
ln t1
1 1
0
+ .
ij =
(3.30)
ln
1 1
0
ln t2
2i

This subsequently leads to the expectation values

n=2 2N
N1
N2
t1 = t 2 =
,

(3.31)
26 We could actually choose and to differ by an integer. This corresponds to shifting the angle associated with
1
2
one stack of branes by 2 relative to the other stack. For simplicity, we take the angles identical so 1 = 2 .
315
where n=2 is the RG-invariant combination of (v0 ) and the cutoff v0

2N 2i(v0 )
2N
.
n=2 v0 e
(3.32)
We see a posteriori that this result is valid in the regime n=2 / 1 and further corrections
amount to an expansion in this parameter.
3.4. A1 theory with cubic superpotentialIIA/M
We now turn to the description of this system in the IIA/M picture. The relevant brane configuration here consists of two NS5s extended along the quadratic curves

2
2
w = g v
(3.33)
4
and separated along x 6 with N1 (N2 ) D4-branes suspended in between at v = /2 (/2). This
configuration is depicted in Fig. 8. The corresponding M-theory lift is described by a genus one
holomorphic curve with boundary conditions

2
w g v 2
(3.34)
as v
4
and embedding coordinate s satisfying27

ds = 2iN i ,
(3.35)
Ai
ds = 2i(v0 ).
(3.36)
Bi
As in the genus zero case, an explicit representation is well known but we shall seek a parametric
description here. Such an approach may be unfamiliar, so we shall discuss it at length.
Because the desired curve has genus one, we shall parametrize it by a single complex variable
z subject to the identifications z z + 1 and z z + where is the modulus of the torus. In all
that follows, we will focus only on the fundamental parallelogram, which is depicted in Fig. 9(a).
As in the genus zero case, we must specify two marked points a1 and a2 on the parallelogram as
the preimages of the points at infinity on the NS5-branes. Unlike the previous example, though,
the embedding is no longer one-to-one at infinity. This is due to the quadratic curving of the
NS5s and leads to the constraint that, while v(z) has single poles at a1 and a2 , w(z) necessarily
has double poles at these points. The embedding coordinate s(z), on the other hand, is multivalued on the z-plane with a cut connecting a1 and a2 and monodromies consistent with (3.35)
and (3.36).
With the analytic structure of v(z), w(z), and s(z) at hand we can in principle proceed to
write them down. To do this in practice, though, we need the analog of the functions 1 and
ln which allowed us to introduce poles and cuts in the previous example. A convenient choice
27 As in IIB (see footnote 26), we could choose to be different by an integer for B and B cycles. This would
1
2
correspond to nontrivial wrapping of M5 along x 10 when one goes around the compact cycle B1 B2 .
316
Fig. 8. NS5/D4 configuration obtained by applying T -duality to the geometry (2.3) with cubic W (v) (3.22) and D5-branes
wrapping the conifold singularities at v = /2.
(a)
(b)
Fig. 9. Parametrizations of the hyperelliptic curve (3.43). (a) Hyperelliptic curve (3.43) with A and B cycles indicated.
(b) Fundamental parallelogram on z plane used to parametrize the curve (3.43) with A and B cycles indicated.
of building blocks for constructing genus one curves is based on the function28
F (z) = ln (z ),
(3.37)
where
(z) =

n=
2 +2inz
ein
1
( + 1).
2
(3.38)
Because (z) has a simple zero at z = , we see that F (z) ln z near z = 0. This means that
we can use F (z) to introduce branch points and derivatives of F (z) to introduce poles. In what
follows, we shall adopt the notation
n
(n)
F (z ai ).
Fi =
(3.39)
z
Detailed properties of these functions and their relation to Weierstrass elliptic functions can be
(n)
found in Appendix C. The most important feature to keep in mind is that Fi introduces an nth
order pole at the point ai . It is also worth noting here that the Fi(n) are elliptic for n > 1 and have
28 This collection of building blocks was recently used by [69] for essentially the same purpose as ours.
317
the following monodromies for n = 0, 1

Fi (z + 1) = Fi (z),
Fi (z + ) = Fi (z) + i 2i(z ai ),
(1)
(1)
Fi (z + 1) = Fi (z),
Fi(1) (z + ) = Fi(1) (z) 2i.
(3.40)
With our building blocks handy, we are now ready to begin writing general expressions for
the embedding functions. We start with v(z) and w(z), whose analytic structure was described
above. If we add the requirement that w v 2 near a1 and a2 , the most general possibility, up
to constant shifts, is given by29
(1)

(1)
v = X F1 F2 F (1) (a) i ,
(2)
(2)
w = C F1 F 2 ,
(3.41)
where we have inserted the constant shift F (1) (a) i so that w(v) behaves like (3.34) and a is
the separation between marked points
a a 2 a1 .
(3.42)
As described in Appendix C, it is possible to work out the polynomial relation between w and v
explicitly. It takes the form of a hyperelliptic curve
w 2 = P2 (v)2 f1 v f0
(3.43)
with P2 (v) a particular quadratic polynomial in v from which we can read off relations between
the curve parameters X and C of (3.41) and the physical parameters g and of (3.34)
C
,
2 = 12X 2 (a).
(3.44)
X2
The function (z) appearing here is the Weierstrass -function.
We have thus seen that a holomorphic curve with the desired analytic properties along w and
v corresponds to a hyperelliptic curve of the form (3.43) and admits the parametric representation (3.41). We have also found a convenient way to parametrize the moduli space of such curves,
as they depend on two complex parameters, and a. These are analogous to the quantities Si on
the IIB side as they encode essentially the same information.
Let us finally turn our attention to the embedding coordinate s, which is a multivalued function of z satisfying (3.35) and (3.36). To determine its form, we must identify those cycles on
the z-plane that correspond to our A and B cycles. Illustrations of both representations of the
hyperelliptic curve (3.43) which identify all the relevant cycles can be found in Fig. 9. Imposing
the A-periods (3.35) uniquely fixes the form of s up to an integration constant, for which we
make a convenient choice30,31
g=
s = (N1 + N2 )(F1 F2 ia) + 2iN1 (z A),
(3.45)
29 The relative sign of F (1) and F (1) in v is fixed by requiring v to be elliptic while the relative sign of F (2) and F (2)
1
2
1
2
in w is obtained by requiring w v 2 at a1 and a2 .
30 Our choice of integration constant simplifies the limit used to obtain the local geometry near one of the D4 stacks.
It also renders our curve invariant under a particular Z2 symmetry of the parametrization z z, a1 a2 ,
a2 a1 , s s.
31 The relative coefficient of F and F is fixed by requiring ds to be elliptic.
1
2
318
where we defined
a 1 + a2
A
(3.46)
.
2
We can now fix the moduli and a by imposing the B-period constraints (3.36). First note
that equivalence of the two noncompact B-periods implies that

ds = 0,
(3.47)
B2 B1
or in other words
s(z + ) = s(z).
(3.48)
This leads immediately to a relation between a and

N1
.
N
Evaluating the noncompact B-periods then leads to the further condition

2

v
2iN1 N2
,
+
2i(v0 ) = ds = N L(a, ) + ln 02
N

a=
(3.49)
(3.50)
B1
where

12 (a)(a )2
L(a, ) ln
.
( )2
(3.51)
The conditions (3.49) and (3.50) are our final result for the constraints on moduli of the M5
curve.
3.4.1. Comparison with IIB
As in our genus zero example, there are several ways to compare with IIB. The most direct is
to compute the period matrix ij of the elliptic curve (3.41) explicitly in terms of the parameters
a and . For this, we find

2

v0
1
1 1
a 0
ij =
(3.52)
L(a, ) + ln
+
.
1 1
0
a
2i
2
It is now easy to verify that the condition (3.28) on the moduli for supersymmetric vacua in IIB
is solved exactly when (3.49) and (3.50) are satisfied.
If we are interested in computing the Si , though, these can also be determined by direct integration as discussed in Appendix C. In principle, they can be computed exactly as functions of a
and , though the expressions are somewhat complicated and hence not very enlightening. A natural limit to study is that of Im 1 which, based on the identification of cycles in Fig. 9(a),
we naively expect to be identified with a limit of large separation . Indeed, we can justify this
expectation by noting that, in this regime, the ratio n=2 / becomes

n=2 2N
e2iN1 N2 /N + ,
(3.53)

319
where we have suppressed terms that are further exponentially suppressed at large Im . Expanding the Si in this limit as well, we find

n=2 2N
N1
N2
+ ,
t1 = t2 =
(3.54)

where the ti are as in (3.29). This agrees with the IIB result (3.31) that was obtained in the
same regime. Because the moduli of the M5 curve exactly solve the vacuum equation (3.28),
this agreement will persist to all orders in the parameter n=2 /. Though it was not our main
objective, it is nice that the elliptic function formalism leads to exact results for the moduli Si
as obtaining them from the IIB side requires a full solution of the DijkgraafVafa matrix model
[7072] (see also [69]).
3.4.2. Reliability of the M5 and NS5 descriptions
Finally, let us address the question of when our M5 and NS5 interpretations of the curve of
(3.41) and (3.45) are reliable without the use of BPS arguments. For this, we can borrow many
of the results of Section 3.2.2 and Appendix A provided n=2 / is sufficiently large. In this
case, we note that the effective size S of each tube is captured by the corresponding Si while the
effective mass, obtained by expanding the superpotential (3.22) near one of its critical points, is
given by m = g. As usual, the conditions for a reliable M5 interpretation at strong coupling
are straightforward but a bit complicated. These can be easily worked out from Eq. (A.20) in
Appendix A.
The conditions for a reliable NS5 interpretation at weak coupling, on the other hand, take a
fairly simple form

Si
Si
3 3 S
gs 1,
gs N min
,
g
, gSi ,
1 min
i . (3.55)
g
g 3 3
It is not difficult to see that these conditions can indeed be satisfied. Consider, for instance, the
behavior of (3.55) at leading order in n=2 /

gs N
n=2 N/Ni
gs 1,
min(1, g),

1 n=2 N/Ni
1
(3.56)
min 1, g 3 3 .
g

These are easily seen to hold for a wide range of parameters. This condition will be of interest to
us later when studying configurations with D4s and D4s.
4. The brane/antibrane system
We now proceed with our study of IIA/M configurations obtained by applying T -duality to
the brane/antibrane system of [16]. After first reviewing the IIB construction of [16], we will
discuss the NS5/D4 configuration and its M theory lift.
4.1. Branes and antibranes on local CY in IIB
In the recent paper [16], it was suggested that interesting SUSY-breaking configurations could
be constructed by wrapping D5s and D5s at singular points of local CalabiYau. In particular,
320
if the singular S 2 s wrapped by branes and antibranes are homologous, the absence of a conservation law preventing their eventual annihilation suggests that this system is metastable and will
eventually decay.
Because supersymmetry is broken, it might seem that obtaining any quantitative information
about the system is out of the question. However, it was suggested in [16] that the breaking
is sufficiently soft that one retains a significant amount of computational control. In particular,
they conjectured that large N duality continues to hold, permitting one to replace the branes and
antibranes with fluxes on a deformed geometry. Then, it was further argued that the N = 2 SUSY
of IIB strings on this deformed geometry is broken spontaneously by the fluxes, leading one to
expect that essential aspects of the physics continue to be captured by the N = 2 prepotential
which, in turn, is determined by special geometry.
The simplest example of such a system has been considered at length in [16,21] and consists
of wrapping N1 D5s and N2 D5s at the singular points of the A1 fibration (2.3) with cubic superpotential (3.22). After the geometric transition, we are left with the deformed geometry (3.23),
repeated here for convenience

2 2
x 2 + y 2 + w2 = g 2 v 2
(4.1)
f 1 v f0
4
and 3-form fluxes

H = 4 2 N i ,
Ai

H = 4 2 i .
(4.2)
Bi
We will use N i to refer to flux numbers, which can be negative, and Ni to the number of branes
or antibranes so that
N1 = N 1 ,
N2 = N 2 > 0.
(4.3)
We also take 1 = 2 = for simplicity as in the supersymmetric example of the previous section.
The presence of the fluxes (4.2) leads to generation of the superpotential
WGVW N i i + i S i ,
(4.4)
where as usual we define

1
1
F
Si
,
i
H = i,
2i
2i
S
Ai
ij =
Bi
i
2F
= i j.
j
S
S S
(4.5)
The absence of solutions to the supersymmetric vacuum equation (3.28)32

+ ij N j = 0
(4.6)
explicitly demonstrates that SUSY is broken and necessitates a computation of the scalar potential for the lowest components of S i to determine the actual vacuum. Because the breaking is
spontaneous, it is natural to conjecture, as the authors of [16] did, that the Khler potential of the
32 That no solutions exist follows from the fact that Im is positive definite.
321
system continues to be determined by special geometry

Kij S k = Im ij S k
(4.7)
and hence that the scalar potential is given by

V = + ij N j (Im )1j k + kl N l .
(4.8)
The moduli S i are now determined by minimizing this quantity. This was studied by the authors
of [16] who found that, to leading order in the S i , the minimization equations can be written in
the relatively simple form
Re() + Re(ij )N j = 0,

Im() + Im(ij )N j = 0
and lead to the expectation values
2N1 2N2
v0
v0
2N2 ,
t1N1 = t2N2 =
e2i(v0 ) = 2(N1 +N2 ) 2N1
(4.9)
(4.10)
where the ti are as in (3.29). In the above, we have implicitly defined the RG-invariant scale
2(N1 +N2 ) = v02N1 v02N2 e2i(v0 ) .
(4.11)
As in the supersymmetric example of the previous section, we see that this approximation is valid
for small |/|.
Subsequent corrections can in principle be determined using the Dijkgraaf
Vafa matrix model technology [7072]. This was recently done in [21] and an interesting phase
structure uncovered. We will see later how this structure arises in the IIA/M analysis.
4.2. Brane/antibrane configurations in type IIA and M theory
We now consider the brane/antibrane configuration of [16] from the IIA/M point of view.
Applying T -duality, we find a pair of quadratically curved NS5-branes extended along the curves

2
w = g v 2
(4.12)
4
and separated along x 6 . There are also N1 D4s suspended between the NS5s at v = /2 and
N2 D4s at v = /2. This configuration is depicted in Fig. 10.
This classical brane configuration fails to capture any quantum effects of the system and,
strictly speaking, is valid only for gs = 0. We expect that, just as in the supersymmetric examples
discussed previously, quantum corrections will smooth out the NS5/D4 and NS5/D4 intersection
points and also lead to a dimpling of the NS5s. To study this process, we propose to follow
the procedure adopted before and lift this configuration to M theory. At large gs , this system is
described by an M5-brane wrapping a smooth minimal area surface in wvs space. At small gs ,
this M5 is more appropriately viewed as a curved NS5-brane with flux.
As usual, we shall work in a regime where the M5 (NS5) worldvolume theory is reliably
approximated by the NambuGoto action (or its descendant)
SNG =
(4.13)
g
322
which, when reduced along the flat 0123 directions, looks similar to the worldsheet action
of the bosonic string [56]. Standard analysis of this action indicates that a given surface is an
extremum provided the embedding functions are harmonic on the worldsheet
v(z)
= w(z)
= s(z)
=0
(4.14)
and satisfy a Virasoro-type constraint33

gs2 s s + v v + w w = 0.
(4.16)
Notice that holomorphic embeddings automatically satisfy both conditions. In our situation,
though, we do not expect to find a holomorphic curve because the presence of antibranes signals a breaking of supersymmetry.
Before proceeding let us note a few technical points. First, in order to prevent the open string
tachyon from destabilizing our system, we need the separation between D4 and D4 stacks to
to be small when necessary, where

satisfy > 1. Second, in what follows we will also take
is as in (4.11). Note that such a condition is not very stringent because, as we shall see later in
Section 4.5, if we make the physical assumption that |v0 | > then all solutions corresponding
to true minima have this property.
4.2.1. First attempts at an M-theory lift
For illustrative purposes, let us begin by attempting to construct a holomorphic lift of the
configuration in Fig. 10 to see what goes wrong. The first step along this direction is to address
the boundary conditions along w and v

2
2
w g v
(4.17)
as v .
4
We have already seen how to deal with these in the previous section. In particular, we saw that
the resulting wv geometry must be a hyperelliptic curve which admits a parametric description
of the form (3.41)

v = X F1(1) F2(1) F (1) (a) i ,
(2)
(2)
w = C F 1 F2
(4.18)
with g and determined as in (3.44)
C
,
2 = 12X 2 (a).
X2
Our only task, then, is to write a holomorphic 1-form ds with the appropriate periods

1
1
i
ds = N ,
ds = i .
2i
2i
g=
Ai
(4.19)
(4.20)
Bi
33 We have assumed here a target-space metric of the form
ds 2 = gs2 |ds|2 + |dv|2 + |dw|2 + ,
(4.15)
where has been set to 1 as usual and the manifest gs -dependence has been introduced in accordance with the factor of
R 1 in (2.21).
323
Fig. 10. NS5/D4/D4 configuration T -dual to the brane/antibrane system of [16].
The condition on the A-periods uniquely fixes34

s = (N1 N2 )(F1 F2 ia) + 2iN1 (z A),
(4.21)
where A is defined as in Eq. (3.46). If we now try to impose the B-period constraints, though,
we find a problem. In particular, because 1 = 2 = we must have

(4.22)
ds = 0
B2 B1
which in turn implies that

a a 2 a1 =
N1
> .
N 1 N2
(4.23)
This is impossible because both a1 and a2 lie within the fundamental parallelogram by assumption. This result is not surprising. It simply illustrates that the obstruction to finding a
holomorphic lift of the configuration in Fig. 10 is the lack of a well-defined embedding coordinate s that yields the appropriate wrappings along x 10 .
This situation is easily improved, though, if we take away the constraint of holomorphy and
instead simply require that s be harmonic, continuing to satisfy one of the minimal area conditions (4.14). In this case, we can write it as the sum of a holomorphic and an antiholomorphic
function and it is easy to find a two-parameter family of such with the right A-periods35
s = (N1 N2 + )(F1 F2 ia) + i(N1 + )(z A)
+ i(N1 )(z A).

+ (F1 F2 + i a)
(4.24)
We now have enough freedom to fix the B-periods to whatever we like without saying anything
about a and . It is thus generically possible to write down a harmonic embedding s(z, z ) with
all the desired properties for arbitrary values of the moduli.36
34 As in the supersymmetric example of Section 3.4, s is fixed only up to an integration constant for which we make a
particularly convenient choice.
35 We have again made a convenient choice for the integration constant in s.
36 More generally, a holomorphic 1-form on a genus g curve of this type will have g free parameters, which is enough
freedom to fix the A-periods but further imposing constraints on B-periods fixes the moduli. On the other hand, a harmonic 1-form on a genus g curve of this type will have 2g free parameters, which is enough freedom to fix all periods
for arbitrary values of the moduli.
324
It may seem that, with the moduli unfixed, we have too much freedom and indeed this is the
case. With the introduction of nonholomorphic contributions to s, we are now faced with the
daunting task of addressing the nonlinear constraint (4.16), which is no longer satisfied. As we
shall see, this will select a particular s(z, z ) from the two-parameter family (4.24) and, in so
doing, combine with the B-period constraints to fix the moduli. For now, however, let us be very
naive and try to use physical reasoning to pick a particular s, postponing a further discussion of
(4.16) to the next subsection.
At large separation , which we argued in the previous section corresponds to large Im( ), we
expect the curve to roughly factorize into a holomorphic piece, describing the local geometry
near the D4s, and an antiholomorphic piece, describing the local geometry near the D4s. A realistic expectation, then, is that if we write s(z, z ) as the sum of holomorphic and antiholomorphic
parts
s(z, z ) = sH (z) + sA (z)
(4.25)
then the periods of dsH will reflect only the contribution from the D4s and the periods of dsA
only the contribution from the D4s when Im( ) is large. Imposing this condition picks out the
choice = N2 , = N1 in (4.24)
s = N1 (F1 F2 ia) + 2iN1 (z A) + N2 (F1 F2 + i a).
(4.26)
Further imposing the B-period constraints for this particular s then leads to the following conditions on a and
N1 = N1 a N2 a,
2

2

v
v0
2i(v0 ) = N1 L(a, ) + ln
+ N2 L(a, ) + ln 02
+ 2iN2 a.
2

(4.27)
Note the rough similarity of these equations to the vacuum equations (4.9) obtained on the IIB
side at leading order in /.

We can make the comparison more explicit by using the expression
(3.52) for the period matrix ij of the hyperelliptic curve (4.18) to write (4.9) as

2

v
N1 N2
N1 ( a)
1
+
L(a, ) + ln 02
,
0 = Re
1
N2 a
2i

2

v0
N1 + N 2
N1 ( a)
1
+
0 = Im
(4.28)
L(a, ) + ln
.
1
N2 a
2i
2
It is now easy to see that (4.27) and (4.28) are equivalent and hence that the curve (4.18) with
embedding coordinate (4.26) satisfying (3.35) and (3.36) has the same moduli as the nonsuper
symmetric IIB vacuum at leading order in /.
4.2.2. A few problems
Though our physically-motivated curve (4.18), (4.26) enjoys some success when comparing
to IIB, two important problems remain. First, the connection with IIB is valid only at leading
order in /
and fails to account for any further corrections. The second problem, which is not
unrelated to the first, is that our curve fails to satisfy the additional constraint (4.16) and hence is
not a true minimal area surface.
In fact, from (4.16) we see that the introduction of any nonholomorphic dependence to s
necessitates further nonholomorphic contributions to v and w. This further alters the geometry
325
and, in particular, makes it impossible to write down an M-theory lift of the configuration in
Fig. 10 with a holomorphic relation between w and v. As a result, we cannot hope to obtain a
geometry that is T -dual to the local CY (4.1) with fluxes except in an approximate sense.
How then can we explain the nice agreement with IIB produced by our physically-motivated
curve (4.18), (4.26)? Because it cannot be an exact solution, the best we can hope for is that, as
alluded to above, it is an approximate one. To make this more precise, note that, at leading order
in /,
the characteristic size of contributions to (4.16) from the embedding coordinate s is gs N
while those from v and w are 2 / and g 2 , respectively.37 This means that the nonholomorphic contributions v, w to v, w typically scale like
v
(gs N )2
,
2
(gs N )2
,
g 2
and hence are suppressed relative to the holomorphic ones precisely when
2
gs N
min(1, g).

(4.29)
(4.30)
Quite nicely, this is less stringent than the second of the conditions (3.56) required for our analysis based on the NambuGoto action to yield a reliable weakly-coupled IIA description of the
configuration as a curved NS5-brane with flux.38 In other words, when a IIA interpretation is
reliable the nonholomorphic corrections to w and v are always suppressed!
This does not help at all with the problem of identifying the right s from the 2-parameter
family of possibilities (4.24), though. As we shall see, there are a couple of ways to deal with
this. The most direct is to actually find an exact solution to (4.16) and study it in the limit of
Eq. (4.30). This will be done in the next subsection. We could alternatively try to discern the
behavior of the curve in this regime directly from the action without actually finding an exact
solution. This will be addressed in Section 4.4.
4.3. An exact M-theory lift
Following the discussion above, we are led to conjecture that the configuration depicted in
Fig. 10 can be lifted to an M5 curve that, in the limit (4.30), is approximated by a holomorphic
geometry of the form (4.18) with a suitable harmonic embedding coordinate s. To test this conjecture, let us first search for an exact solution. This is most easily accomplished if the number
of D4-branes, N1 , is equal to the number of D4-branes, N2
N1 = N 2 N
(4.31)
because the additional symmetry leads to significant simplifications.

As described in Appendix D, the resulting curve satisfies
Re( ) = 0,
= 2a,
(4.32)
37 The easiest way to see this is by mapping the parameters ,

, and g to the effective quantities n=1 and m which
determine the geometry near either brane stack. In particular, n=1 = 2 / and m = g. We now use the fact that the
genus 0 curve has characteristic scales v n=1 and w mn=1 to obtain the desired result. One can also obtain the
scaling by studying the elliptic functions in (3.41) directly, but this is less transparent.
38 The second equation in (3.56) gives different conditions depending on the values of N , but it is least stringent for
1,2
N1 = N2 , for which it is the same as (4.30).
326
Fig. 11. Tilting of D4s and D4s that gives rise to logarithmic bending along the w-direction in the M-theory lift (4.33).
which causes a number of elliptic quantities to vanish and permits us to write a relatively-simple
exact solution

i
+ i(z A) + c.c. + iN(z A + z A),

s = Nr0 cos F1 F2
2

2(gs N )2 (1)

(1)
(1)
(1)
F1 F2 F (1) (a) + i ,
v = X F1 F2 F (1) (a) i +
X

i
gs N (2)
(2)
w = gs N r0 sin F2 F1
i(z A) + c.c. +
F1 F 2 ,
2
r0 sin
(4.33)
where
3 2 (/2)
2
(4.34)
,
3 (/2)2 g2
6 (/2)2 2g2
and g2 is one of the Weierstrass elliptic invariants as defined in Appendix C. This solution as
written admits three free parameters, X, , and . The first two are analogous to the quantities
X and C in the supersymmetric curve (3.41), (3.45) and encode the boundary conditions of
the curve along w and v. The third, , is determined in terms of the boundary data by the
noncompact B-period of ds.
There are a number of interesting features to this solution but let us focus for now on one in
particular, namely that the functions Fi , which exhibit logarithmic behavior and have described
the bending of NS5s along x 6 in our supersymmetric examples, make their appearance in w.39
The reason for this is quite simple to understand. The branes and antibranes can lower their
energy by moving closer together. Because of the NS5s, though, this requires a rotation into the
w direction and carries an energy cost associated to the corresponding increase in length. In the
equilibrium configuration, the D4s and D4s are thus rotated a bit, changing the direction along
which they pull on and dimple the NS5s as illustrated in Fig. 11. This is directly reflected in
the exact solution (4.33), where indicates the rotation angle.
r02
4.3.1. The IIA regime

Let us now turn to the regime in which the configuration (4.33) can be reliably interpreted as
a curved NS5-brane with flux. From the discussion of Section 3.4.2, we see that this corresponds
39 Note that F , F , and their conjugates enter in a combination that is single-valued in a manner analogous to the
1
2
function ln ||2 .
327
to the regime (3.56)

gs 1,
gs N

2
min(1, g),
g
2

min 1, (g)3 .
(4.35)
We now look for solutions with approximately holomorphic boundary conditions along w and v
of the form

2
2
+ ,
w(v) g v
(4.36)
4
where ellipses are used to indicate the possibility of nonholomorphic terms which we have previously argued must be suppressed when the second condition of (4.35) is satisfied. To do this, let
us rewrite (4.33) in the form of (4.18) with nonholomorphic corrections. Using (3.44) to relate
curve parameters to the physical ones g and , we obtain

4(gs N r0 )2
i
+ i(z A) + c.c.
s = Nr0 1
F1 F2
2
g 2 4
+ iN(z A + z A),

(1)

(1)
F1 F2 F (1) (a) i
v=
12 (/2)

gs Nr0 2 (1)
(1)
+4
F1 F2 F (1) (a) + i
,
||

2 gs Nr0 2
i
w=
+ i(z A) + c.c.
F2 F1 +
g

2
2

g
(2)
(2)
(4.37)
+
F F2 .
12 (/2) 1
From this, it is quite easy to see that the nonholomorphic contribution to v is negligible provided40
gs N
(4.38)
1.

To determine precisely when the nonholomorphic contribution to w is negligible, though, one
(n)
must look a bit more closely at the properties of the elliptic functions Fi . The largest contributions to the ratio w/w come from the regions near the D4 and D4 tubes and hence near the
(n)
2.
midpoints between a1 and a2 . There, it is not difficult to show that the Fi all scale like (/)
1
2
41
This means that the first term of the expression for w in (4.37) scales like g (gs N/)
while
the second term scales like g 2 . This means that the nonholomorphic contribution to w is suppressed when
gs N
g

(4.39)
40 Note that we use the fact that r 1, which is true when /
is small as we are assuming.

0
41 Note in particular that the contribution from the F is suppressed relative to that of the constant term (z A).
i
328
and hence that our solution reduces to a hyperelliptic curve along w and v with harmonic embedding coordinate s when
gs N
min(1, g ).
(4.40)

Note that this condition is less stringent than what we expected from general reasoning (4.30).
The reason for this is that while s as a whole scales like gs N , the division into holomorphic
and antiholomorphic pieces is not symmetric. In particular, the holomorphic part of s scales like
gs N but the antiholomorphic piece actually goes instead like gs N 2 /2 < gs N . This means
that gs N can actually be a bit larger than our previous reasoning would have suggested. We
expect this behavior to persist even for more general numbers of branes and antibranes so that
nonholomorphic contributions to w and v can be neglected for a parameter range that is wider
than what we need for the IIA interpretation to be reliable.
4.3.2. The reduced curve
Let us now focus on the regime (4.40) in which the nonholomorphic corrections to w and v
in (4.37) can be dropped and we are left with the approximate solution

i
+ i(z A) + c.c. + iN(z A + z A),

s = r0 N F1 F2
2
(1)

(1)
F1 F2 F (1) (a) i ,
v=
12 (/2)
g2 (2)
(2)
F 1 F2 .
w=
(4.41)
12 (/2)
This geometry looks precisely like its supersymmetric counterpart (3.41) with the holomorphic
embedding along s (3.45) simply replaced by a harmonic one. Note that a very specific expression
for s has emerged with an extra factor of r0 that we were unable to produce without the exact
solution (4.33). As expected, this reduces to our guess (4.26) when Im( ) is large because in this
regime r0 becomes
r02 = 1 + 40ei + .
(4.42)
Without the solution (4.41) in hand, though, we were unable to say anything about possible
subleading corrections away from the limit of infinite Im( ). Requiring the existence of an exact
solution has now completely fixed the ambiguity of (4.24) for all values of .
This geometry is connected by T -duality to a local CY of the form (4.1) with fluxes just as
the large N duality conjecture of [16] would suggest. But what about the moduli? We saw in the
previous section that they agree in the limit Im( ) but, armed with the correction terms
in r0 , can we go further?
Quite nicely, the answer to this question is a resounding yes. The complex structure of the
curve (4.41) is completely fixed by the relations
Re( ) = 0,
= 2a,

2
2

v
v
+ L(/2, ) + ln 02
i ,
2i(v0 ) = r0 N L(/2, ) + ln 02
(4.43)
(4.44)
where L(a, ) is as in (3.51) and the condition (4.44) arises from the B-period constraint
in (4.20).
329
As discussed in Appendix D, it is now easy to verify, using the expression (3.52) for the period
matrix ij of the geometry (4.41), that a and satisfying (4.43) exactly solve the equations of
motion which follow from the IIB potential (4.8). In the regime where our exact solution (4.33)
provides a reliable IIA description of the system in terms of a curved NS5-brane with flux, it is
thus exactly T -dual to the IIB system after the large N transition to all orders in /!
4.4. Direct analysis of the NS5 action
In order to better understand this agreement, we now turn to a direct study of the M5/NS5
action in the IIA regime (4.35) where the M5-brane can be described as an NS5-brane with
flux.42 We shall find that, in this limit, the IIB potential (4.8) makes a natural appearance and is
responsible for fixing the moduli from the IIA/M point of view as well.
In M theory the M5-brane was curved in the v, w, s = R 1 (x 6 + ix 10 ) directions. Upon
reduction to IIA we expect a single NS5-brane curved in v, w, x 6 . The winding of the original
M5-brane around the M-theory circle x 10 is encoded in the nontrivial configuration of the oneform field strength Fm of the worldvolume theory of the NS5-brane in IIA. The relevant part of
the NS5-brane Lagrangian is [12,68]:

1
I= 2
(4.45)
det gmn + gs2 Fm Fn .
gs
The NS5-brane configuration is specified by imposing the boundary conditions
w(v) Wn (v)
(4.46)
at infinity, requiring the fluxes of Fm to be consistent with the numbers of D4s and D4s that we
started with and fixing the logarithmic bending of x 6 at infinity.
Because the boundary conditions (4.46) along v and w are independent of gs and N , the
corresponding embedding functions can remain macroscopically large regardless of the value of
gs N . This means that, provided we take gs N to be sufficiently small, it is possible to obtain a
parametric separation of scales relevant for the w, v and s parts of the geometry, respectively.
Indeed, the second condition defining the IIA regime (4.35) is essentially just this because it
is equivalent to imposing |ds/dw|, |ds/dv| 1.43 As a result, it is natural to separate the
contribution to the induced metric coming from x 6 and write the NS5 worldvolume action
as:

1
+ x 6 x 6 + g2F F ,
I= 2
(4.47)
det gmn
m
n
s m n
gs
where the induced metric g is that associated to the embedding coordinates w and v, treating x 6 as a constant. As explained above, the last two terms scale like (gs N )2 . So defining:
ds =
dx 6
+ iF
gs
and expanding to the first nontrivial order in s we find

42 The worldvolume action has previously been used to obtain Khler potentials in [73].
43 Recall that, on general grounds, we expect |ds| g N , |dv| 2 /, |dw| g .
(4.48)
330
1
I= 2
gs

1

g +
ds ds,
2
(4.49)
where is with respect to the metric g .

Because we do not have to worry about the complications associated with finding exact solutions to the equations of motion, let us now be completely general here and try to minimize
(4.49), starting with a configuration of NS5s wrapping holomorphic curves
w 2 (v) = Wn (v)2
(4.50)
for Wn (v) a polynomial of degree n and suspending arbitrary numbers of D4s and D4s between
them at the n critical points of Wn (v). These configurations are related by T -duality to the general
ones of [16] with many conifold singularities. As we argued above we can replace the NS5s and
D4/D4-branes with a single NS5-brane with fluxes turned on. As usual, we shall denote the RR
fluxes associated to these branes by N i with negative fluxes corresponding to antibranes. The
embedding of the NS5-brane has to satisfy the boundary conditions:
w(v) Wn (v)
(4.51)
at infinity and the periods of ds must be consistent with the fluxes as in (2.22) and (2.23)

ds = 2iN i ,
ds = 2i.
(4.52)
Ai
Bj
If we consider only the first term of (4.49), ignoring the second for a moment, we find that the
embedding described by w and v alone must be of minimal area. Since we impose holomorphic
boundary conditions (4.51) at infinity, the entire surface must be holomorphic and we are again
led to the family of hyperelliptic curves (2.12)
w 2 = W (v)2 fn1 (v).
(4.53)
Note that the complex structure moduli of these curves are true flat directions of the first term
in (4.49).
If we now consider only the second term in the action and minimize it with respect to s for
given and fixed embedding w(v), the equations of motion imply that ds is a harmonic 1-form
with respect to the induced metric g . As we shall see explicitly in a moment, there is a unique
harmonic 1-form ds with specified A and B periods for each value of the complex structure
moduli of the curve (4.53).
This gives us our embedding but what about the moduli? Despite the fact that they correspond
to flat directions of the first term in (4.49), this is not true of the second s-dependent term. The
moduli must therefore be chosen to minimize this quantity which, as we shall see, is nothing
more than the IIB potential (4.8).44
4.4.1. Construction of ds and the effective potential
To construct the harmonic form ds with the desired periods we proceed in the following way:
we can parametrize the complex structure of the surface (4.53) by the periods of 12 w dv on the
44 Consequently, we see that the two terms in (4.49) are, in a sense, not quite decoupled.
A cycles:
1
2i
1
w dv = Si .
2
331
(4.54)
Ai
The periods of w dv on the B cycles are determined by the holomorphic prepotential F of the
curve:

1
1
F
(4.55)
.
w dv =
2i
2
Si
Bi
For each value of Si we want to construct a harmonic form ds on the surface with the desired
periods. A harmonic form can be written as the sum of holomorphic and antiholomorphic forms.
It is convenient to pick the following basis of holomorphic forms:
i =
1 w
dv
4i Si
as they have nice periods on the A cycles:

j = ij
(4.56)
(4.57)
Ai
and on the B cycles:

2F
j =
= ij ,
Si Sj
(4.58)
Bi
where ij is the period matrix of the surface. The harmonic form ds is the sum of holomorphic
and antiholomorphic forms:
ds = hi i + i i .
(4.59)
The 2n constraints (4.52) arising from imposing the specific A and B periods now determine
the 2n coefficients hi and j completely in terms of and the period matrix in a manner that is
N , and recalling that
,
straightforward to determine. Collecting hi , i , N i , i into vectors h,
the i are all identical, we can write the result in a compact form

h = + 2i N,
= (Im )1 (
+ N ).
(4.60)
To fix the moduli, we must now minimize (4.49). We focus on the second term because this is the
only one that depends on the moduli. Using (4.60), we can write it directly in terms of and

1
V=
ds ds
(2)2

1
+ T [ ]
= 2 Im hT [ h]
2

= (
+ N)(Im
)1 ( + N ) 2 Im T N .
(4.61)
332
This is precisely the effective potential on the IIB side obtained from special geometry [16] including the zero-point shift introduced therein. Consequently, we see that when we can reliably
use the NambuGoto action to provide a reliable description of the system at weak coupling, the
IIA picture quite generally reproduces the IIB story complete with some aspects of the off-shell
physics. Note that from this point of view, the supersymmetric vacua also fall out nicely because
any holomorphic ds that satisfies the constraints (4.52) automatically minimizes (4.49).
4.4.2. More on the connection with IIB
We can make the above agreement of the effective potentials computed in the IIB and IIA
systems more transparent as follows. In IIB after the geometric transition we have a noncompact
CalabiYau with fluxes. Without flux, the system would have flat directions parametrized by the
complex structure moduli of the CalabiYau. The fluxes then create a potential for these moduli
which stabilizes them. This potential (4.8) can be written suggestively as
VIIB = Kij i W j W ,
(4.62)
where W is the GVW superpotential (2.8) and Kij the Khler metric (4.7). Using standard results
from rigid special geometry, it is not difficult to show that the scalar potential (4.62) is simply
the electromagnetic energy of the H flux of the system

VIIB = Kij i W j W = H H .
(4.63)
CY
Now the analogy with IIA should be clear. There, we started with two NS5-branes on the
curves w W (v) which get replaced by a single curved NS5 on a hyperelliptic curve of
the form (2.12)
w 2 = Wn (v)2 fn1 (v)
(4.64)
x6,
with 1-form flux Fm on its worldvolume turned on and logarithmic bending along
both of
which are combined into our complex coordinate s. Without the flux and accompanying bending,
the system has flat directions parametrized by the complex structure moduli of the hyperelliptic
curve. The presence of nontrivial ds induces an effective potential for the moduli and lifts the
degeneracy. The form of this potential is simply

VIIA ds ds
(4.65)
which, given the identification (2.30)

ds H
(4.66)
S2
is nicely consistent with T -duality. Indeed, the tension of our bent NS5 with flux in IIA is directly
identified with the energy stored in the nontrivial H -flux in IIB.
4.5. When are our solutions true minima?
Before closing this section, let us come back to a subtle point that needs to be addressed.
While we have demonstrated that our curves solve the equations of motion that follow from the
IIB potential (4.8), we have not yet said anything about the nature of such solutions or even
333
demonstrated that they exist in the first place. In particular, the moduli are fixed by (4.43) but
inverting this equation to actually determine is nontrivial.
Precisely this issue was addressed on the IIB side in the recent paper [21], which used the
DijkgraafVafa matrix model technology to compute subleading corrections to the potential (4.8)
at large Im( ) and study the structure of its critical points. They found a fairly intricate phase
structure that should make an appearance in our formalism as well. In this section, we briefly
comment further on this. While far from our original goal, one nice benefit of our work is that
we have exact expressions in hand for both the solutions as well as the IIB potential for a cubic
superpotential and equal numbers of branes and antibranes. This allows us to obtain exact results
for many of the quantities which characterize the phase structure of [21]. It is important to point
out, though, that the primary simplification arises not from a particular property of the IIA/M
description itself but rather from the formalism we developed to study it,45 which does not easily
generalize to curves with higher genus.
Before proceeding, let us first revisit the manner in which moduli are fixed from the IIA/M
point of view. As described in detail in Appendix D, the conditions (4.43) follow directly from
requiring that s be periodic along the compact B-cycle and imposing a Z2 symmetry on the full
solution. The former is a necessary condition for existence of a well-defined curve46 while the
latter is required because we imposed boundary conditions that preserve the obvious Z2 of the
brane/antibrane system. Consequently, from the IIA/M point of view, simply having a smooth
curve with the right boundary conditions already fixes all of the moduli except for the imaginary
part of , which we hereafter refer to as T
T Im .
(4.67)
This quantity is finally fixed in terms of the boundary data (v0 ) by imposing the noncompact
B-period constraint in (4.20). The result is (4.44)
4
v0
2i(v0 ) = r0 N 2L(iT /2, iT ) + T + ln ,

(4.68)
where we have implicitly used the fact that L(iT /2, iT ) is real. As we saw in the previous section, imposing this condition is equivalent to further extremizing the approximate action (4.49)
with respect to T . The form of this action is equivalent to the IIB potential (4.8) and takes the
form

(2i/N)2
N2
+ T .
V (T ) =
(4.69)
2[L(iT /2, iT ) + T ] + ln | v0 |4
In the IIA/M, approach, we are thus naturally led to study the potential (4.69) as a function of a
single real parameter T , which is easily seen to correspond essentially to the parameter in [21].
Critical points of this potential correspond to solutions of the full equations of motion. Because
the expression (4.69) is exact for all T 0, we should in principle be able to study the nature of
its critical points in quantitative detail even away from the regime T 1.
45 In particular, it is the parametrization of the complex moduli space by a and that makes things simple. This does
not come without a price, though, because the dictionary relating these quantities to the Si (C.39) is quite complicated.
46 Of course, we could have taken s to be periodic along the compact B-cycle up to an integer multiple of 2 R , but
10
we have chosen this integer to be zero for simplicity.
334
As a first use of (4.69), we can evaluate the potential at the degeneration point T = 0. In [21],
it was argued that this quantity is independent of N1 and indeed this can be verified explicitly
V (0) =
2(i)2
L(0, 0) + 12 ln | v0 |4
(4.70)
In fact, this is precisely the form of Eq. (103) of [21] with Bt = L(0, 0).
It is also fairly straightforward to study the qualitative structure of the critical points of (4.69),
which correspond to solutions of (4.68). To do so, we note that
2L(iT /2, iT ) + T
(4.71)
satisfying47
is a monotonically increasing function of T

2L(0, 0) = 6 ln 2,
lim L(iT /2, iT ) + T =
T
(4.73)
while r0 is a monotonically decreasing function of T satisfying

lim r0 = ,
T 0
lim r0 = 1.
(4.74)
Consequently, if we fix |v0 /| and view (4.68) as defining as a function of T , the situation is
as depicted in Fig. 12 with 2i attaining a minimum value 2i at the point T = T . What we
see from this is that for 2i > 2i there are two branches of solutions. At 2i , these two
branches merge so that for 2i < 2i there are no solutions at all. From this structure, we
can already conclude that one of the branches describes local minima and the other local maxima
of (4.69). Physically, we expect that the T > T branch yields the minima and indeed it is easy
to see that this is the case by looking at the derivative of (4.69),

N2
(2i/N)2
T V (T ) =
(4.75)
1 2
r0 (2L(iT /2, iT ) + T + ln | v0 |4 )2
which of course vanishes when (4.68) is satisfied. Because T V (T ) is positive at both T = 0 and
T = it follows that, whenever V (T ) has two critical points, the one at larger (smaller) T is
a local minimum (maximum), in agreement our expectations and the results of [21]. Also note
that, depending on the size of 2i, solutions along the T > T branch may have higher or lower
energy than the point T = 0 (4.70) and hence do not always correspond to global minima of the
potential. Sample plots of V (T ) that exhibit all of these features are presented in Fig. 13.
What we have described above is precisely the phase structure of [21], to which we refer the
reader for a more thorough discussion. The critical point 2i corresponding to a particular
choice of |v0 /| is determined by the equations
4
v0
4(g2 3 (iT /2)2 )
ln =
(4.76)
2L(iT /2, iT ) T ,

6 (iT /2)(3 (iT /2) + 21 ) g2
47 Because of these properties, we must have that |v | > ||/2 2 in order prevent V (t) from diverging and becoming
0
negative at small t . The need for such

a condition is clear as we do not expect to be able to take the cutoff smaller than
the physical scale ||. The factor of 2 2 can be understood by noting that the curve (3.41) becomes

w2 = g 2 v 2 v 2 2 /2
(4.72)
at t = 0. The cuts extend from / 2 to 0 and are centered at /2 2. The above condition simply corresponds to the
sensible requirement that the physical cutoff to be larger than this scale.
335
Fig. 12. Solutions to the constraint (4.68). The T > 0 branch corresponds to local minima of (4.69) while the T < 0
branch corresponds to local maxima.
(a)
(b)
(c)
Fig. 13. V (T ) for various choices of 2 i. (a) 2 i > 2 i and sufficiently large that the local minimum is also a
global minimum of V (T ). (b) 2 i > 2 i but sufficiently small that the local minimum is not a global minimum of
V (T ). (c) 2 i < 2 i so there is no local minimum and hence no solution to (4.68).
2i
4r0 (g2 3 (iT /2)2 )
.
=
N
6 (iT /2)(3 (iT /2) + 21 ) g2
(4.77)
Unfortunately, these expressions are quite complicated so, even though we can in principle solve
for T and 2i in full generality, it is difficult to do so in practice without resorting to numerics.
As noted in [21], though, one should be able to move all of the interesting physics to the regime
of large T , where (4.76) and (4.77) become more manageable, by tuning |v0 /| appropriately. To
see this from (4.76) and (4.77), note that the LHSs of both equations are monotonically increasing functions of T and hence that T and 2i both grow as |v0 /| is increased. This means
that we can always ensure that T 1 by choosing |v0 /| to be sufficiently large. Expanding
(4.76) in this regime, we find
4
v0
eT
T + O(1)
ln =
(4.78)

20
which, using (C.40), is equivalent at large T to the condition for minimizing Eq. (87) of [21].
From this, we conclude that for |v0 | ||
4
4

v0
v0
T

= 20 ln + 20 ln 20 ln + .
e
(4.79)

336
Turning now to Eq. (4.77), we can expand at large T

2i eT
=
+ O(1)
N1
20
and subsequently apply (4.79) to obtain Eq. (90) of [21]
4 4
v0 v 0
2i
= 20 ln + .
N1

(4.80)
(4.81)
Another regime in which (4.76) and (4.77) simplify is that of small Im( ) but physically this is
not very interesting because the description that we have in hand breaks down there. As discussed
in [21], additional degrees of freedom that have not been accounted for enter the story. On the
IIB side, D3- and D5-branes wrapping the shrinking compact B cycle become important. On the
IIA side, these correspond to D4- and D6-branes with boundaries on the curved NS5 which wrap
the compact B-cycle.
Of course, there is much more to say about the phase structure of this system than what we
have described here. For this, we refer the reader to the discussion of [21], which can easily be
translated to the IIA/M picture via T -duality.
Finally, to address a comment made at the beginning of Section 4.2, let us note that from
(4.76) we can determine numerically that, if we take |v0 | > , then T 32 and hence all true
4 at most of O(102 ).
minima have ei (/)
5. Discussion
In this section, we comment on various lessons that can be drawn from our IIA/M analysis.
First, in Section 5.1 we shall address the possible conclusions one can reach about the nature
of supersymmetry breaking in these setups with particular attention paid to when some residual
supersymmetric structure remains. Following this, we turn in Section 5.2 to the implications that
observations made in [12] have on our configurations, how they arise in the T -dual type IIB setup,
and the general lessons one can draw from them for studies of metastable nonsupersymmetric
configurations in local contexts. Finally, in Section 5.3 we address some open questions that
would be interesting to study further.
5.1. Brane/antibrane configurations and spontaneous supersymmetry breaking
Though we have described our type IIA brane/antibrane configurations as being T -dual to the
IIB setups studied in [16], there is an important point that must be emphasized. All of our analysis
has relied on the fact that the circle on which we perform T -duality is large in the IIA picture.
On the other hand, the analysis of [16] implicitly assumes that this circle is large on the IIB side
so, in reality, the two approaches are focusing on quite different regions of parameter space. In
supersymmetric situations, this typically is not important because BPS arguments guarantee that
quantities such as the compact complex structure moduli in IIB are protected as the radius is
varied. In our case, though, supersymmetry is broken so our ability to connect the IIB and IIA
pictures at weak coupling indicates that some additional structure must be present there.
Indeed, the analysis of [16] relies quite heavily on the argument presented therein that, after
the geometric transition, supersymmetry is spontaneously broken in these setups, at least at string
tree level. This additional structure is what ensures that the Khler potential determined from
337
special geometry is reliable and can be used to compute the potential (4.8) that fixes the moduli.
It must also be responsible for our successful comparison of the IIB and IIA pictures.
In a sense, then, our ability to reproduce the results of [16] in weakly coupled IIA despite the
fact that we have essentially tuned the IIB radius to zero provides further evidence in support of
the assertion made there that the breaking of supersymmetry is spontaneous, at least for some
range of parameters. Moreover, we may be gaining a hint as to what parameters actually control
the severity of supersymmetry breaking because the novel features of our solutions are con Whenever both of these quantities are small, the
trolled by the objects gs N/ and gs N/g.
minimal area curve takes the form (4.41) and can be successfully compared to type IIB. On the
other hand, when either is large, new phenomena such as the introduction of nonholomorphic
bending along w are introduced.48
Indeed, the very structure of our curve suggests that supersymmetry is spontaneously broken when gs N/ and gs N/g are small because there it factorizes into two pieces which
are separately holomorphic but with respect to different complex structures. In other words, the
solution factorizes in this regime into two pieces which are individually supersymmetric but preserve different sets of supercharges. It is only the simultaneous presence of both which leads
to breaking of supersymmetry. This nice structure strongly suggests to us that supersymmetry
thus remains spontaneously broken, and hence our solution remains reliable, even outside of the
regimes discussed in Appendix A provided gs N/ and gs N/g are both small.
A natural question to ask now is whether or not the relevance of gs N/ and gs N/g can
be understood more directly. At strong coupling, their importance is rather obvious because, as
we have seen, they control the backreaction of the nonsupersymmetric fluxes, reflected by the
nonholomorphicity of s, on the w, v geometry. When they become important, nonholomorphicity
enters into w and v and destroys the nice factorized form alluded to above. Of course, we could
have said essentially the same thing about our curved NS5 brane with flux at weak coupling.
A more difficult question, though, is how these parameters appear in the open string picture
with the D4- and D4-branes. There, it is well known that the presence of the D4s leads, at string
tree level, to the introduction of an FI D-term which spontaneously breaks supersymmetry.49
Because there is no structure to prevent it, we expect that the breaking becomes more severe
when string interactions are taken into account. One might therefore expect that the breaking
remains spontaneous provided gs 1. From our exact solution, though, we see that it is also
necessary for gs N to be bounded in some sense. What, though, is the physical relevance of gs N ?
For this, consider an open string amplitude, given by inserting open string vertex operators,
corresponding to fields living on D-branes, on a worldsheet (see Fig. 14). Let the worldsheet be
a sphere with h 1 holes and n 0 handles added. There are h boundaries on the worldsheet.
Insert ki open string vertex operators
on the ith boundary, where i = 1, . . . , h. Let the total
number of vertex operators be k = hi=1 ki . The amplitude then goes like
gok C gok go4+2h+4n = gok4+2h+4n .
(5.1)
1/2
gs
is the normalization constant for open string vertex operators and C is the
Here go
amplitude without insertions.50 This leads to the term in the five-dimensional action on the
48 Note that we can reliably turn on g N/ and/or g N/g while continuing to use our minimal area curve for a
s
s
range of parameters at strong coupling gs 1 so they are indeed real and can be studied reliably in the M-theory regime.
49 For a review of this, see for instance [74].
50 Recall that C g 4 , and adding a hole to a worldsheet gives a factor of g 2 while adding a handle gives a factor
o
o
S2
of go4 ; this is how we got the above power of go .
338
Fig. 14. Worldsheet with n 0 handles and h 1 holes. represents an open string vertex operator.
D4/D4-branes:

gok4+2h+4n
d 5x
h

tr X ki ,
(5.2)
i=1
where we denoted the fields collectively by X, which can be the adjoint or bifundamentals.
To get the conventional normalization where the kinetic terms have 1/go2 in front of them, let
us rescale fields as X X/go . Then we have
go4+2h+4n

d 5x

h

tr(X ki )
tr X ki = N h go4+2h+4n d 5 x
.
N
i
(5.3)
i=1
If we integrate over x 6 , this becomes

N h go4+2h+4n b

4
d x
h

tr(X ki )
i=1
N 2 gs2n (gs N )h1

4
d x
h

tr(X ki )
i=1
(5.4)
where we defined a four-dimensional t Hooft coupling constant by

4
go2 N
gs N
.
b
b
(5.5)
From (5.4), it is clear that gs counts the number of handles and gs N counts the number of
boundaries. Thus, we see that even if gs is small we can still get sizeable stringy corrections
from gs N provided N is large enough.51 Determining precisely how large gs N must be, though,
is a more difficult question which requires a knowledge of how and appear in the various
interaction terms (5.4). It would be very interesting to pursue this further and understand the
relevance of the specific quantities gs N/ and gs N/g from this open string point of view.
51 It was noted in [75] that, in the g 0 limit with fixed N , the gauge theories geometrically engineered by D-branes
s
on CY singularities receive contributions only from disks and thus have only single-trace operators. Here we are refining
and generalizing this argument by introducing parameters gs , gs N , and 4 .
339
Fig. 15. Decay process of nonsupersymmetric metastable brane/antibrane configuration. The displayed bending of NS5s
is for the x 6 direction; the curving in the wv direction is not shown. a) Initial metastable configuration. D4 and D4 are
pulling NS5s inward. b) D4 and D4 have pair annihilated by quantum tunneling. c) Without tension, NS5s straighten.
d) Straightening propagates to infinity.
5.2. Metastability and boundary conditions

We have constructed nonsupersymmetric configurations in type IIA/M theory by suspending
D4s and D4s between curved NS5s. Because of their tension, D4s and D4s pull and bend the
NS5s logarithmically in a manner that extends to infinity (Fig. 15(a)).
D4s and D4s being separated by a potential barrier, this configuration is classically stable, but
in quantum theory they can pair annihilate by quantum tunneling, as we have already discussed
in the introduction.52,53 The time it takes for this tunneling process to occur can be estimated
by a standard analysis, considering an instanton interpolating the initial and final configurations
[16]. Note that, for the physics of this tunneling, the fact that the NS5s are bent logarithmically
does not matter; the D4 and D4 know only about their vicinity and what is happening far away
is irrelevant for this instantaneous process. This is very much as the -decay of a nucleus in an
electric field; the decay rate, governed by quantum tunneling, is not affected by the electric field
outside the nucleus. Once all the D4s and D4s have pair annihilated (Fig. 15(b)), the tension
that was holding NS5s is no more and the NS5s start to straighten (Fig. 15(c)). This part of
the decay process proceeds according to the classical equation of motion, much as the fact that,
once the -decay has taken place, the motion of the emitted positron in the electric field can be
treated classically. The straightening of the NS5s propagates outward (Fig. 15(c), (d)), but the
time it takes for the straightening to reach the cutoff v = v0 goes to infinity as one takes v0 .
Namely, in the v0 limit, the system has a runaway instability. Note that the energy released
in this process becomes infinite as v0 , because not only the mass of the D4/D4 but also the
tension of the infinitely long bent NS5 gets converted into energy. Although here we explained
the decay process in the IIA language for the sake of argument, the decay process in M theory is
exactly the same.
This is in contrast with what happens if there is no logarithmic bending of NS5s, which is the
case if gs is strictly zero: gs = 0. In that case, there is no straightening part of the above decay
process, since the NS5s are straight from the beginning (in the x 6 direction; in the wv directions
they are holomorphically curved). So, the decay process ends in a finite time when the D4s
and D4s have gone through quantum tunneling and pair annihilated. The final configuration is a
52 For simplicity of the argument, we are assuming that the numbers of D4s and D4s are the same, N = N = N .
1
2
53 A similar analysis of different configurations appeared in [23].
340
supersymmetric configuration of two NS5s along the holomorphic curves,

w = W (v).
(5.6)
However, for gs = 0, which is the case we have been focusing on, there is logarithmic
bending. Therefore, for gs = 0, our nonsupersymmetric D4/D4 system and the supersymmetric system (5.6) have different boundary conditions for NS5s at infinity. In general, to properly
define the quantum theory of noncompact systems, one must specify the boundary conditions at
infinity. This means that, the metastable configuration of Fig. 15(a) and the supersymmetric configuration (5.6) are not different states in the same theory but rather different states in different
theories. Therefore, the former can never decay into the latter, which was the emphasis of [12].
Note that this does not mean that the configuration of Fig. 15(a) is stable. As we have seen,
it does decay and shows a runaway instability. This is not at all in contradiction with [12] but
actually [12] gives a nice interpretation of this runaway instability as follows: although our nonsupersymmetric system wants to decay into the supersymmetric configuration (5.6), it can only
keep decaying forever toward the latter, because the latter does not lie in the same theory. The
fact that the energy released in the process is infinite also implies that this process cannot end in
a finite time.
In the above, we used the fact that the D4/D4 pull and bend the NS5s to explain that the
boundary conditions are different for nonsupersymmetric and supersymmetric configurations.
An alternative way to understand this is the following. A D4-brane ending on an NS5 is a vortex
charge for the 1-form field strength Fm on the NS5 worldvolume theory. Because D4 and D4 have
opposite vortex charges, the total charge on each NS5 is N N = 0. This is the same as for the
supersymmetric configuration (5.6), which has no D4/D4 and hence no vortex charge. However,
D4-branes ending on an NS5-brane act also as charges for the x 6 scalar field, with respect to
which D4 and D4 have the same charge. Therefore, the total charge is N + N = 2N and is
different from that of the supersymmetric configuration, which is zero. So, the two configurations
have different charges and one cannot decay into the other.
The boundary condition being different for nonsupersymmetric and supersymmetric configurations can be understood in IIB also. If we wrap D5s on 2-cycles in CY and turn on B-field
through them, D3 charges are also induced. A careful study of the system of H3 , F3 , F5 field
strengths shows that in the presence of these sources, the field B2NS has to grow logarithmically
with v, once we include the leading-order backreaction of the D5s on the geometry.54 So, while
at tree level we can specify the B2 flux through the 2-cycles to be equal to a constant (which is
the T -dual statement of specifying the x 6 position of the NS5-brane to be constant), once we add
D5s and consider their backreaction, we see that B2 grows logarithmically at infinity, necessarily
modifying the boundary conditions of the system. Now the difference of boundary conditions is
easy to see: if we start with N D5s and N D5s, the logarithmic divergence of B2 at infinity will
be proportional to 2N . On the other hand, in the supersymmetric configuration, B2 is constant
and there is no logarithmic running. So the boundary conditions at infinity are different. Also
the energy difference between the two configurations diverges logarithmically if we integrate the
energy of the H field over the entire space.
Finally, as we saw in Section 4, the boundary condition difference in M theory is more serious. Besides the different logarithmic bending of the x 6 coordinate, we also have to change the
asymptotic behavior on the w, v plane to solve the minimal area equations.
54 The easiest way to see the logarithmic running of B is from the picture after the geometric transition. The running
2
of B2 is supposed to capture the RG flow running of the gauge coupling.
341
5.3. Future directions

In the present paper, we studied a class of nonsupersymmetric metastable configurations and
found that they can be analyzed in a reliable way using the NS5/M5 worldvolume actions in
various regimes of type IIA/M theory.
We focused on the case of two D4/D4 stacks between two quadratically bent NS5-branes
in type IIA or, in the T -dual IIB description, local CY geometry obtained by A1 fibration.
However, the method of nonholomorphic M5/NS5 curves is applicable to more complicated configurations. For example, flavors can be incorporated by including semi-infinite D4/D4-branes
attached to NS5s or by including D6-branes [54]. In particular, such systems should give us an
M-theory viewpoint to look at the nonsupersymmetric metastable states in gauge theory which
were found in [1] and whose possible string theory embeddings were pursued in [10,12]. It is also
straightforward to generalize our method to ADE quiver gauge theories obtained by stretching D4/D4-branes between multiple NS5s, which is T -dual [50,51] to the local CY geometries
obtained by ADE fibrations [44,45]. What is particularly interesting about such more general systems is that, nonsupersymmetric brane/antibrane configurations can be stable rather than
metastable [53]. In such cases it may be possible to take the gauge theory limit and one may be
able to study the protection of holomorphic quantities using gauge theory techniques.
Obtaining exact M5/NS5 curves in such more complicated systems will be increasingly difficult in practice, if not impossible. However, the fact that for small gs N the curve decomposes
into the holomorphic part (wv) and the harmonic part (s) must still hold, and we furthermore
expect that the curve is protected in the same parameter regime because we conjecture that supersymmetry is softly broken there. This may provide a tractable window in which one can study
the vast nonsupersymmetric landscape of string/M theory in a controlled way. Needless to say,
for such a program, it is highly desirable to better understand the relation between the parameter
gs N/ and the softness of the supersymmetry breaking, for which we could give only an indirect
argument in Section 5.1.
Nonsupersymmetric configurations such as the ones studied in the current paper may potentially serve as useful modules or building blocks for constructing realistic phenomenological
and/or cosmological models [29]. Of course, for that it is crucial to study if one can realize
such configurations in compact CYs as metastable configurations. For example, in the IIA brane
configurations or its dual IIB local CY geometries, the superpotential W (v) is given by hand
through the boundary condition at the cutoff v = v0 . If we were to embed these models in compact CYs, those boundary conditions must arise dynamically from some mechanism in the rest
of the CY beyond the cutoff v = v0 .
Acknowledgements
We would like to thank D. Berenstein, J. de Boer, J. Heckman, N. Iizuka, T. Okuda, H. Ooguri,
Y. Ookouchi, D. Robbins, J. Seo, E. Sokatchev, C. Vafa, M. Van Raamsdonk and especially
A. Sen for discussions. J.M. is grateful to the theory groups at the University of Texas, UCSB, and
KITP for their hospitality during the course of this work. J.M. and M.S. would also like to thank
the theory group at Harvard University for hospitality while this work was being completed.
The work of J.M. and M.S. was supported in part by Department of Energy grant DE-FG0392ER40701. J.M. is also supported by a John A. McCone postdoctoral fellowship and M.S.
by a Sherman Fairchild Foundation postdoctoral fellowship. The work of K.P. is supported by
Foundation of Fundamental Research on Matter (FOM).
342
Appendix A. Validity of M5/NS5 descriptions55

A.1. Validity of M5/NS5 worldvolume actions
In M theory in 11D spacetime, which is not compactified or is compactified on a circle of
radius much larger than the 11D Planck length 11 , the worldvolume dynamics of M5-brane is
described by a NambuGoto action SM5 at energy scales much lower than 1
11 . This description
becomes inappropriate if the M5 worldvolume starts to have structures smaller than 11 . For
example, worldvolume with extrinsic curvature R 2
11 is bad, and two M5s within distance
11 is bad.
If we compactify M theory on a small circle of radius R, we know that there is a weakly
coupled dual description: type IIA string theory where F1 string is the lightest object and perturbation theory in gs 1 is valid [76]. The relations between M theory and IIA quantities are
R = gs s ,
1/3
11 = gs s .
(A.1)
In type IIA, we have NS5-brane, whose worldvolume action SNS5 we know to be of NambuGoto
type as given in (4.45) [68]. This description becomes inappropriate if the NS5 worldvolume
starts to have structures smaller than the string length s (note that the 10D Planck length 10
1/4
gs s is smaller than s ). Worldvolume fluxes with sufficiently large density is bad too, which
we will discuss later. If we start from M theory with gs 1 and go to IIA with gs 1, there
is transition of these pictures at gs 1 which is highly nontrivial. Although we do know that
the worldvolume action for M5 is NambuGoto for gs 1, we do not know a priori how these
actions get corrected when we go through this transition at gs 1. Indeed, when we compactify
M theory on a circle of radius R = gs s and lower gs , the distance between images of M5-branes,
1/3
which is gs s , will be of the same order as 11 = gs s when gs 1, and we have no control
over dynamics there. What saves the day is supersymmetry, which fixes the action up to twoderivative order. Thanks to this, we can safely say that the action must be NambuGoto all the
way through the transition at the lowest order. It was logically possible that light objects, which
make the M5 action unreliable at gs 1 at scales 11 , become much lighter than the string
mass Ms = 1
s after the transition and invalidate the use of NS5 NambuGoto action for gs 1.
But we know a posteriori that this actually does not happen, since in IIA the lightest objects are
strings, and as long as we are below the energy scale 1
s the NambuGoto action for NS5 does
not get corrected. Because of strong coupling, it is in general impossible to follow how the mass
of those light objects changes as one changes gs in order to check that the potentially dangerous
light objects in M theory indeed become heavy and safe in IIA after transition. However, for
BPS objects this is possible because their mass does not get renormalized. For consistency, the
mass of such BPS objects must continue to be large even after the transition. One possible object
is the M2-brane stretching between image M5-branes (Fig. 16(a)). To determine the BPS mass,
we can safely use the M5/M2 picture at gs 1, since then the distance between M5s is larger
than 11 . From a 10D point of view, this looks like a string and its tension is T = RTM2 R 3
11 .
If we decrease gs , we cannot use the M5/M2 picture any more, but this object continues to
exist and its tension is still given by T . When we get to IIA, where gs 1, the tension T is
T = (gs s )(1/gs 3s ) = 1/ 2s . So, this object has tension of order IIA string tension, and is not
dangerous as far as low energy ( 1
s ) physics is concerned. Now consider the validity of the
55 We are grateful to Ashoke Sen for a helpful discussion on this subject.
(a)
343
(b)
Fig. 16. Some features of M-theory lifts of NS5 branes with and without flux. (a) M2 stretching between image M5s.
(b) M-theory lift of NS5 with flux.
NS5 worldvolume action (4.45) in the presence of flux. Let there be n fluxes per length 2 s so
that the flux density is n/2 s . The M-theory lift is as shown in Fig. 16(a). Although tilted, the
situation is the same as in Fig. 16(b) in the covering space, except that the distance between two
M5s is now

2gs s
2gs s , ngs 1,
=
(A.2)
2 s
ngs 1.
2
n ,
1 + (ngs )
The tension of the BPS object coming from M2 stretching between M5s is
TF1
TM2
,
1 + (ngs )2
(A.3)
where TF1 = 1/2 2s is the string tension. Because now this is the lightest object in IIA, its
tension must be at least of order of string tension. This is the case if ngs 1. Therefore, the
NambuGoto action (4.45) for NS5 is valid if
ngs 1.
(A.4)
Note that gs 1 is implied since we are in IIA.

A.2. Validity of M5/NS5 curves
Let us consider when the curves obtained by using the NambuGoto action to study M5/NS5
setups are reliable. Consider the simplest case of A1 with quadratic superpotential studied in
Section 3.2. Even in more general cases, near each throat the curve can be approximated by this.
The M5/NS5 curve is, from Eq. (3.14),

1
1
s = N log ,
v= +
(A.5)
,
w = m
.
Here we redefined in (3.14) and also shifted s by a constant. is related to a in (3.14) by

2 = a. Note that we are setting = 1. For the simplicity of the argument, let us set henceforth
m = 1.
(A.6)
344
Fig. 17. The M5 curve in the throat region, for fixed x 6 . As we go once around the circle in the vw plane, we go N
times along the M-theory circle. is the distance between different strands of M5.
A.2.1. Validity as NS5 curve

First, consider the validity of (A.5) as an NS5 curve. The thinnest part of the throat, where we
expect that the flux is densest and the curvature is largest, is at || = 1. If we set = ei ,
s = iN,
v = 2 cos ,
w = 2i sin .
(A.7)
Therefore, as changes from 0 to 2 , we go once around the circle of radius on the

Re v Im w plane and N times along the M-theory circle (Fig. 17). So, the flux density is
n N/. From (A.4), we obtain the condition for the NS5 worldvolume action to be valid:
gs N
.
(A.8)
Another condition comes from requiring that the extrinsic curvature R of the NS5 curve be
small in string units. One can readily compute the extrinsic curvature of the curve (A.5), which
is equal to the curvature of the induced metric on the -plane56 :
x 1,
R=
2||4 [x 2 (||4 + 1) + 8||2 ]

.
2 [x 2 ||2 + 2(||4 + 1)]3
(A.9)
It is easy to see that the maximum of this function is as follows:
4
x 2 3: R = 2 2
at ||2 = 1,
(x + 4)2

4x 6
x 4 48 (x 4 48)2 64x 4
2
x 2 3: R =
at || =
.
272 (x 4 16)2
8x 2
In the present case (A.8), the maximum curvature is
R 2 .
(A.10)
(A.11)
By requiring this to be much smaller than s

1 .
= 1, we obtain another condition:

(A.12)
In summary, for the curve (A.5) to be trustable as an NS5-brane curve in IIA, the following
conditions must be met:
gs 1;
1, gs N .
(A.13)
56 This is the extrinsic curvature for s, v, and w. Strictly speaking, one must compute the curvature for x 6 , v and w, but
this does not make any qualitative difference.
345
The first condition is necessary for the IIA string theory to be valid in the first place. We can
write this also as
gs 1,
1,
x 1.
(A.14)
If we restore m, one can show that (A.13) is replaced by

gs 1,
gs N min(m, 1),
1
min m3 , 1 .
m
(A.15)
A.2.2. Validity as M5 curve

Next, consider the validity of the curve (A.5) as an M5 curve. First of all, we need
gs 1
(A.16)
for the M-theory description to be valid.

We need that distance between two different strands of the M5 curve must be smaller than
1/3
the Plank length 11 = gs . The distance in the throat part of the curve is given by (see Fig. 17)
=
gs
N 2 gs2 + 2
=
1/3
Therefore, to have gs
2/3
gs
gs
x2 + 1
(A.17)
we need

x 2 + 1.
(A.18)
2/3
We also need the curvature (A.9) to be much smaller than 2

11 = gs
(A.10), we obtain
x 2 3:
x2
x 2 3:
everywhere. From
1
1/3
1 gs
,
+4
1
x3
1/3
gs
.
4
(x 16) x
(A.19)
In summary, for the curve (A.5) to be trustable as an M5-brane curve, the following conditions
must be met:
x 1:
2/3
1,
2/3
x, gs
1/3
,
2/3
x.
gs
1/3
gs
,
1/3
x = gs N.
(A.20)
Here we dropped factors such as 2 3 which are inessential as far as the order estimates are concerned. Since gs 1, N 1, some conditions in the above are automatically satisfied. Therefore,
a more economical way to state the condition is
x 1:
gs
x 1:
gs
x 1:
gs
(A.21)
346
Fig. 18. Parameter regimes in which the curve (A.5) is trustable as an NS5/M5 curve, for (a) 1 and (b) 1. The
M5 description is valid in the shadowed regions with M5, while the NS5 description is valid in the shadowed region
with NS5. The shadowed regions mean that the NS5/M5 descriptions are valid far away from the surrounding lines,
not inside the surrounding lines; see (A.14) and (A.20).
A.2.3. Summary
Combining the result (A.14) for IIA and the one (A.20) for M theory, we obtain the parameter
regimes of type IIA/M theory in which the curve (A.5) is trustable as an NS5/M5 curve, as shown
in Fig. 18.
As discussed in the main text, nonsupersymmetric curves simplify in the x 1 limit and
agree with the geometry obtained in type IIB in [16,21]. Not only IIA but also M theory has a
parameter regime in which this condition, x 1, is satisfied, although one must take to be
large too. Note that 1 is also allowed in M theory.
Appendix B. T -duality between IIB 2-forms and IIA NS5 position
In Section 2.3, we discussed the T -duality between TaubNUT space in IIB and the NS5brane configuration in IIA. At the level of the Buscher rule, the NS5-branes are delocalized
(smeared) in the y direction. However, in string theory, the NS5-branes are expected to become
localized. Indeed, it is known that the y position of the NS5-brane is dual to B-field through
certain 2-cycles (which is sometimes called the dyonic coordinate in the literature) in the
TaubNUT geometry [63,64]. Although one could study this localization of NS5-branes using
worldsheet CFT techniques [65,66], here we present an alternative approach to determine the
position of NS5-branes which, to our knowledge, is new. Specifically, we consider wrapping an
imaginary D5-brane around the 2-cycle and follow the T -duality. The D5-brane gets mapped
into a D4-brane stretching between two NS5-branes. From how much the D4-brane is tilted in
the y direction, one can read off the y position of NS5-branes on which the D4-brane ends.
In the TaubNUT geometry (2.26) in IIB, there are nontrivial 2-cycles cpq obtained by fibering
the y circle Sy1 over a segment connecting zp and zq in K. The harmonic self-dual 2-form pq
dual to cpq can be locally written as (see, e.g., [64,77]):
1
pq = d(p q ),
2
p = H 1 Hp (d y + ) p ,
dp = 3 dHp .
(B.1)
347
pq is localized near the 2-cycle cpq , and is normalized as

pq = 2R.
(B.2)
cpq
Let us consider turning on the following NSNS B-field:

B = bpq ,
(B.3)
which is a Wilson line for B through cpq . From (B.2), one sees that b has the following
periodicity:
2
(B.4)
.
R
If we T -dualize the IIB metric (2.26) in the presence of the B-field (B.3) using the Buscher rule,
one still obtains exactly the same IIA metric as in (2.27) if we replace the coordinate y with y
defined by
b
=b+
b
y y + (p q )y ,
(B.5)
2
where the subscript y denotes the y component.
Now, consider wrapping an imaginary D5-brane around the 2-cycle cpq (the remaining 3 + 1
worldvolume dimensions extend in the Minkowski space M4 ). On the D5 worldvolume, the Bfield (B.3) is equivalent to the worldvolume gauge field flux
b
2 F B = bpq = d(p q ).
2
The corresponding gauge potential 1-form is
(B.6)
b
2 A = (p q ) b.
2
(B.7)
Here is a closed 1-form which is determined as follows. Since the y circle Sy1 shrinks smoothly

to zero at z = zp , zq , the Wilson line along the y-circle,

S 1 A, must vanish at these points; othery
wise there would be a delta-function like F -flux at these points on the D5,
which is unphysical.
On the other hand, from the explicit form of p in (B.1), one sees that S 1 (p q ) is nonvany
ishing even at z = zp , zq . The 1-form must be chosen to cancel this Wilson line57 :

Hp Hq
Hp Hq
Hp Hq
=d
y =
d y + y
i
dzi .
2H
2H
2H
(B.8)
If we T -dualize along y,
we obtain a D4 stretching between two NS5-branes at z = zp , zq , which
lies along the curve
b
y(z) = 2 Ay = (p q )y + by .
2
(B.9)
57 A similar term cannot be added to (B.5) because that would correspond to a gauge transformation nonvanishing at
infinity, which is a global symmetry transformation rather than an immaterial coordinate change.
348
In the y coordinate (B.5), this curve is

Hp Hq
b/2,
z = zp ,
y (z) = by = b
=
b/2, z = zq .
2H
(B.10)
Therefore, the relative y position of NS5s, which are the endpoints of the D4, is given by b.
The periodicity (B.4) is consistent with the radius of the T -dual circle, (2.28).
To summarize, the k-center TaubNUT geometry (2.26) with B-field (B.3) through the 2cycles cpq in IIB is T -dual to the IIA configuration with k NS5-branes, where the pth and qth
NS5-branes are separated in the y direction by b. Similarly, RR 2-form through the 2-cycle cpq ,
C = cpq ,
c
=c+
2
,
R
(B.11)
corresponds to the distance between the NS5-branes in the x 10 direction, when lifted to M
theory.58 In an equation, the relation between the 2-forms in IIB and the relative position of
NS5-branes in IIA is

i
(B.12)
C2RR + IIB B2NS = 2is,
gs
where s = (x 6 + ix 10 )/R and R = gsIIA is the radius of the M-theory circle.

Appendix C. Properties and applications of some elliptic functions
In this appendix, we summarize some of the properties of elliptic functions used throughout
the text.
C.1. Building blocks
We begin by recalling some basic definitions. Our primary building blocks for constructing
genus one curves are based on the function (3.37)
F (z) = ln (z ),
(C.1)
where denotes the Jacobi elliptic function denoted elsewhere by 3 or 00 :

(z) =
2 +2inz
ein
n=
1
= ( + 1).
2
(C.2)
It is easy to see that (z) has a simple zero at z = and hence that F (z) ln z near z = 0. By
taking derivatives, we can introduce functions with poles at the points ai
n
(n)
Fi =
F (z ai ).
(C.3)
z
(0)
We also typically denote Fi
= F (z ai ) simply by Fi .
58 This can be seen by noting that taking S-dual in IIB corresponds to switching the two S 1 s of M theory on T 2 . In the
present case, T 2 is spanned by x 10 and y . In the type IIA picture, the x 10 coordinate appears as a periodic scalar field
living on the NS5 worldvolume.
349
When building our genus 1 curves, it is important to know the periodicity properties of the Fi .
These are easy to work out from the periodicities of (z)
(z + ) = ei 2iz (z).
(z + 1) = (z),
(C.4)
This leads to the nontrivial periods (3.40)

Fi (z + 1) = Fi (z),
Fi (z + ) = Fi (z) 2i(z ai ) + i,
(1)
(1)
Fi (z + 1) = Fi (z),
Fi(1) (z + ) = Fi(1) (z) 2i
(C.5)
as well as
(n)
(n)
(n)
Fi (z + ) = Fi (z + 1) = Fi (z)
(C.6)
for n > 1.
The Fi(n) also have nice properties under z z

F (z) = F (z) 2iz + i,
F (1) (z) = F (z) + 2i,
F (n) (z) = (1)n F (z),
n > 1.
(C.7)
It is now a relatively simple matter to expand any elliptic function, say f (z), in terms of the Fi(n) .
(n)
To do so, we need only identify a linear combination of the Fi with the same pole structure
as f (z). By holomorphy, these two quantities are equivalent up to a constant that is trivially
computed.
(n)
This permits us to relate our collection of functions, Fi , to the more commonly used Weierstrass elliptic functions, whose properties we now review.
C.2. Weierstrass elliptic functions
The three primary Weierstrass elliptic functions are defined by59

z
z2
z
(z) = z
+
exp
,
1
m + n
2(m + n )2 m + n
m,n=
(m,n)=(0,0)
1
(z) = +
z
m,n=
(m,n)=(0,0)
1
(z) = 2 +
z

m,n=
(m,n)=(0,0)

1
1
z
+
+
,
z (m + n )
(m + n )2 m + n

1
1
2
[z (m + n )]
(m + n )2

(C.8)
59 To compare with the usual literature on Weierstrass functions, note that we have set = 1 and = . We also
1
3
2
2
explicitly indicate 1 , 3 -dependence instead of the usual g2 , g3 .
350
and are related to one another by

(z) =
ln (z)
,
z
(z) =
(z)
.
z
(C.9)
An important fact is that (z) is odd. This implies that (z), as well as its even derivatives, are
even while odd derivatives of (z) are odd.
The analytic structure of the Weierstrass functions on the fundamental parallelogram is as
follows. The function (z) has a simple zero at z = 0
(z) z +
ln (z) ln z +
(C.10)
1
+ .
z2
(C.11)
from which we see that

(z)
1
+ ,
z
(z)
The Weierstrass -function is elliptic, having completely trivial periods. The functions and ,
on the other hand, are quasiperiodic
1
(z + 1) = e21 (z+ 2 ) (z),
(z + ) = e23 (z+ 2 ) (z),

(z + 1) = (z) + 21 ,
(z + ) = (z) + 23 ,
(C.12)
where 1 and 3 are the -function half-period values

= 1 ,
= 3
2
2
(C.13)
and which satisfy the nontrivial identity

1 3 = i.
(C.14)
By comparing analytic structures and periodicities, it is easy to relate the Weierstrass func(n)
tions (C.8) to our building blocks Fi (C.3)
F (z) = ln (z) 1 z2 + iz + ln ( ),
F (1) (z) = (z) + i 21 z,
F (2) (z) = (z) 21 ,
n2
(z),
F (n) (z) =
z
n > 2.
From this we also see that the function L(a, ) defined in (3.51) is related to (z) by

12 (a)(a )2
L(a, ) ln
= ln 12 (a) (a)2 21 a 2 + 2ia.

2
( )
(C.15)
(C.16)
The relations (C.15) and (C.16) prove useful because there are a number of nice identities involving Weierstrass functions that are well known. We will list some of these at the end of this
appendix.
351
C.3. Some simple manipulations

(n)
While we can use analytic structure to relate Weierstrass functions to our Fi , it is also
useful to work out polynomial relations among more complicated elliptic functions. Consider,
for instance, the elliptic function
(1)
(1)
V (z) = F1 F2 .
(C.17)
(n)
The function V (z)2 is also elliptic and admits an expansion as a linear combination of the Fi .
To determine which one, we use the expansion
F (1) (z) =

1
g2 z 3
+ i 21 z
+ O z5
z
60
(C.18)
to study V (z)2 near a2

V (z)2
1
2[F (1) (a) i]
z a2
(z a2 )2

(1)
2

+ F (a) i 2 F (2) (a) + 21 + O(z a2 ),
(C.19)
where, as in the main text,

a a 2 a1 .
(C.20)
From the antisymmetry of V (z), we also know that the even order poles at a1 will have the same
coefficients as (C.19) while the odd order poles will have opposite signs. This permits us to write
(1)
(1) 2
V (z)2 = F1 F2

(2)
(1)
(2)
(1)
= F1 + F2 + 2 F (1) (a) i F1 F2

2

+ F (1) (a) i 2 F (2) (a) + 21 .
(C.21)
C.4. Properties of the curve (3.41)

Manipulations such as this are easy to perform even in more complicated situations. One
example of interest is the polynomial relation between w and v of (3.41)
(1)

(1)
v = X F1 F2 F (1) (a) i ,
(2)
(2)
w = C F1 F 2 .
(C.22)
Because of the antisymmetry of both w and v under a1 a2 it is impossible to write w as a

quadratic polynomial in v. The simplest possibility then is that w 2 may be written as a quartic
polynomial in v. Comparing pole structures, one can quite easily deduce the correct relation
w2 =
where

C2
P2 (v)2 + b1 v + b0 ,
4
X
(C.23)
352

P2 (v) = v 2 + 3X 2 F (2) (a) + 21 ,
b1 = 4X 3 F (3) (a),

2
(2)
5 (4)
4 g2
+ F (a) 2 F (a) + 21 .
b0 = X
6
3
(C.24)
This gives an explicit relation between the deformation parameters b0 , b1 and the quantities a
and . It also permits us to relate the variables of our parametric description to the physical
parameters (3.44) associated with boundary conditions. In particular, at large v the relation (C.23)
becomes

C
C
w 2 P2 (v) = 2 v 2 3X 2 (a) ,
(C.25)
X
X
where we have used
F (2) (z) + 21 = (z).
(C.26)
From this, we can read off the parameters g and (3.44)

C
,
2 = 12X 2 (a).
(C.27)
X2
Finally, we note that this formalism also makes it quite easy to compute period integrals. For
instance, the expectation values of the Si in type IIB (3.3) are mapped to period integrals of the
1-form
1
(C.28)
w dv.
2
For genus one curves, we have two Si which can be computed via

1
1
w dv,
Si =
(C.29)
2i
2
g=
Ai
where the A cycles are as in Fig. 9. We can compute these fairly easily by starting from

1
v(z)
XC (2)
(2) 2
w(z)
dz =
F F2
Si =
(C.30)
4i
z
4i 1
Ai
(n)
and writing the integrand as a linear combination of the Fi

(3)
(2)
(4)
(2)
(2) 2
(4)
(3)
(2)
= a F 1 + F2 + b F 1 F2 + c F 1 + F2
F 1 F2
(1)

(1)
+ d F1 F2 + i + e.
(C.31)
In particular, once the coefficients d and e are known we can immediately write
S1 =
g3
(id + e),
4i(12 (a))3/2
S2 =
g3
(id e).
4i(12 (a))3/2
(C.32)
Actually obtaining the coefficients a, . . . , e in practice is not difficult and we find

1
d = 2F (3) (a),
b = 0,
c = 2 F (2) (a) + 21 ,
a= ,
6
g2
7
e=
412 + 41 F (2) (a) + 3F (2) (a)2 + 2F (3) (a)F (1) (a) + F (4) (a).
12
6
(C.33)
353
In the text, we studied the curve (C.22) in the limit of large Im( ). We can expand S1 and S2 in
this regime60 using the first moduli-fixing relation (3.49)
S1 = g3 e2iN2 /N + ,
S2 = g3 e2iN1 /N .
(C.34)
To write this in terms of UV , we must apply the second moduli-fixing relation (3.50). At large
Im( ), it takes the form
2N 2i(v0 )
2N
= (1)N (2X)2N e2iN1 N2 /N .
UV = v0 e
(C.35)
Noting also that

2 4 2 X 2 +
at large Im( ) we obtain (3.53)

UV 2N
= e2iN1 N2 /N +

(C.36)
(C.37)
which explicitly demonstrates that large Im( ) corresponds to small UV /. Finally, using all
of these results we can express S1 and S2 completely in terms of g, , and UV
2N/N1
2N/N2

3 UV
3 UV
S1 = g
,
S2 = g
,
(C.38)

reproducing (3.54).
A specific case of interest in the main text corresponds to N1 = N2 and = 2a. There, the
exact expressions (C.32) for the Si simplify dramatically

g3
2g2
.
4
(/2)
(/2)
2
S1 = S2 =
(C.39)
1
4i(12 (/2))3/2 3
The first few terms of (C.39) at large Im( ) are easily determined

S1 = S2 = g3 ei 34e2i + 984e3i + .
(C.40)
C.5. q-series and other useful formulae

We now list useful formulae, including q-series, for some of the quantities that arise throughout our computations. In what follows, we define
q = ei
as usual.
We start with the differential equation satisfied by (z)

(z) 2
= 4 (z)3 g2 (z) g3
z
(C.41)
(C.42)
which also implicitly defines the Weierstrass elliptic invariants g2 and g3 .

60 Note that for N < N , the expansion of S is a bit tricky. One must first demonstrate that all terms of the form
1
2
2
e2 iN2 n/N cancel so that e2 iN1 /N is truly the leading contribution.
354
We now list some q-series

kq 2k
2
4 2
,
6
1 q 2k
k=1

k 3 q 2k
4 1
g2 = 20
+ 16
,
15
1 q 2k
1 =
(/2) =
2
3
k=1
k=1
kq k
,
1 + qk

2
(1)n q n ,
(1/2) = 1 + 2

n=1
( ) = 2i 1 +
(2n + 1)(1) q
n n(n+1)
(C.43)
n=1

q 2k
sin(z)
+ 1 z 2 + 4
sin2 (kz),
ln (z) = ln
k(1 q 2k )

k=1
(z) = 21 z + cot(z) + 4

k=1
q 2k
sin(2kz),
1 q 2k
(z) = 21 + 2 csc2 (z) 8 2

kq 2k
cos(2kz).
1 q 2k
(C.44)
k=1
We also note that at large Im( ), L(/2, ) has the nice expansion
L(/2, ) = 20q 262q 2 + .
It is also useful to know the -derivatives of various quantities

ln (z)
1 1
1
g2 z 2
2
=
(z) (z)
+ 21 z (z) 1 ,
2i 2
2
24

1 1
g2 z
(z)
=
(z) + (z) (z)
+ 21 (z) z (z) ,
2i 2
12

1
g2
(z)
2

=
2 (z) + (z) (z)
21 z (z) + 2 (z) ,
2i
3

1
1
g2
2
=
21
,
2i
24
g2
1
=
(6g3 8g2 1 ).
2i
Finally, we list the asymptotic behavior of a few quantities for Im( ) 1
(/2) =
2 2
+ ,
3 2
(C.45)
(C.46)
(C.47)
4 4
+ ,
3 4
2
i
1 = 2 +
+ .
355
g2 =
(C.48)
Appendix D. An exact solution for Nbranes = Nantibranes

In this appendix, we describe the derivation of the solution (4.33) as well as some of its
properties.
D.1. Deriving (4.33)
Our goal is to find an exact solution to the minimal area conditions (4.14) and (4.16)
0 = v(z)
= w(z)
= s(z),
(D.1)
0 = gs2 s s
(D.2)
+ v v + w w
for s having the appropriate periods

1
ds = N i ,
ds = i .
2i
Ai
(D.3)
Bi
Recall that Ni denotes the number of D4s or D4s while N i denotes the RR charges. Here,
we focus on the case of N1 D4s and N2 = N1 D4s so N1 = N 1 = N 2 . Recall also that 1 =
2 = . Solutions to (D.1) are given by s, v, w that are sums of holomorphic and antiholomorphic
functions. We will use this observation to write them as
s(z, z ) = sH (z) + sA (z),
v(z, z ) = vH (z) + vA (z),
w(z, z ) = wH (z) + wA (z).
(D.4)
In this notation, the second constraint (D.2) can be written as

0 = gs2 sH sA + vH vA + wH wA
(D.5)
making manifest that the introduction of an antiholomorphic piece to s necessitates the presence
of nonholomorphic terms in v and w.
As discussed in the main text, the most general harmonic s with A-periods as specified in
(D.3) is given by (4.24) with N1 = N2 N . We write this here as
sH (z) = (F1 F2 ) + i(N + )z,
sA (z) = (F1 F2 ) i(N )z.
(D.6)
Imposing the compact B-period constraint

ds = 0
(D.7)
B2 B1
then leads to
2 Im(a) = Im( ) + iN Re( ).
(D.8)
356
This can be further simplified by noting that switching the sign of x 10 has the same effect as
reversing the fluxes. In equation form, this means that the curve should be invariant under the
operation
s s ,
N N
(D.9)
and implies that the quantities and are real. Combining this with (D.8) we see that
2
(D.10)
Im(a).
Note that there is no condition on Re(a) here. This is a peculiarity of the case Nbranes =
Nantibranes . The final B-period constraint will determine the dependence of the moduli on .
We will postpone the discussion of this until after the exact solution (4.33) is derived.
For v and w, we would like to impose holomorphic boundary conditions of the form (4.17)
but this is impossible because (D.2) implies that at least one of these must have nontrivial nonholomorphic contributions.61 As discussed in the main text, the best we can do is write a curve
whose boundary conditions approach (D.5) in the limit gs N 0. To search for a curve of this
type, we start with the expressions for v and w in (3.41) as a seed
(1)

(1)
vH = X F1 F2 F (1) (a) i ,
(2)
(2)
w H = C F 1 F2
(D.11)
Re( ) = 0,
Im( ) =
and look for holomorphic functions vA and wA , as well as additions to vH and wH , that lead to
a solution of (D.2). Before proceeding, we should get a better understanding of the contribution
to (D.2) from s (D.6). We can write it explicitly as
(1)

(1)
(1)
(1)
sH sA = F1 F2 + i(N + ) F1 F2 i(N )
(D.12)
and use (C.21) to express it as a linear combination of the Fi(n) (C.3). The important point to note
here is that the analytic structure is such that even (odd) poles are even (odd) under the parity
operation a1 a2 . As a result, we should only include additional terms in vH , vA and wH , wA
that introduce contributions of this sort into (D.2). This requirement, combined with the fact that
w and v be periodic on the torus, leads to the following ansatz
sH = (F1 F2 ) + i(N + )z,
sA = (F1 F2 ) i(N )z,

(1)
(1)

(1)
vH = X F1 F2 F (a) i ,
(1)

(1)
vA = F1 F2 F (1) (a) i ,

(2)
2iaz
2iz
(1)
(1)
(2)
+ F1 + F2 +
+ C F1 F2 ,
wH = F1 F2 +

2iaz
2i z
+
.
w A = F 1 F2 +
(D.13)
Note that we had to connect the coefficients of various terms in wH and wA in order to make
sure that the full harmonic function w(z, z ) is periodic. We also relied on the fact that Re( ) = 0
(D.10). In addition to C and X, our ansatz has five parameters , , , , .
61 These contributions cannot vanish at both infinities without having unwanted poles elsewhere.
357
To solve the constraint (D.2), we note that the LHS is an elliptic function and hence is completely characterized by its analytic structure. Indeed, to verify (D.2) we need only check that the
LHS vanishes at a1 or a2 . Cancellation of poles will imply that it is a constant function62 while
actual vanishing at a1 or a2 will guarantee that the value of this constant is zero.
The expressions (D.13) contribute terms with poles of degree four and less to (D.2) so there are
five terms, namely the coefficients of the four poles and the constant term in an expansion near a1 ,
that must be set to zero. This sounds promising because we have introduced five new coefficients.
The system, however, is quite nonlinear and expressions can become quite complicated.
To simplify things a bit further, let us appeal to symmetry yet again. The original IIA configuration of Fig. 10 that we seek to lift is symmetric under the Z2 transformation
v v,
w w,
s s,
N N.
(D.14)
It is clear how such a transformation should be realized in the z-plane as it corresponds to an

exchange of a1 and a2 in (D.13). Requiring that our full M5 curve also possess this symmetry63
implies that we must have
= 0.
(D.15)
Our final simplification comes from studying the highest order poles of (D.2). In particular, the
only remaining contributions to third and fourth order poles are

2

(3)
2ia
(3)
(1)
(1)
F1 F 2 +
+ X F1(2) F2(2) .
C F 1 F2
(D.16)
The fourth order pole can be canceled by choosing

2C
(D.17)
X
but this leaves no freedom left for dealing with the third order pole. This only vanishes if
=
2ia
= F (1) (a) i = (a) 21 a
which we can rewrite using (C.14) as
(a) =
23 a
(a) (/2)
=
.
a
/2
(D.18)
(D.19)
This admits the obvious solution

= 2a.
(D.20)
Using (D.10), we see that this fixes Re(a) = 0 as well as requires that the curve parameters
and satisfy
= .
(D.21)
62 Because the poles at a and a are equivalent up to possible minus signs, we need only check that they cancel at one
1
2
of the two points.

63 Note that we must flip the sign of one of , as well in accordance with (D.10). The choice appropriate for realizing
the parity flip (D.14) is .
358
The condition (D.20) is not all that surprising given the symmetry of the problem. It also leads
to tremendous simplifications in the subsequent analysis because
F (n) (/2) = 0 for n odd.
(D.22)
This is easy to see by combining the periodicities (C.5) and parity properties (C.7).
It is now straightforward to check the remaining terms in the expansion of (D.2) near a1 . There
are three such that we need to vanish, corresponding to the two remaining poles and the constant
term. Miraculously, we can achieve a solution by fixing only the two parameters and , leaving
C and X arbitrary as before. This leads to the result
sH = N r0 cos (F2 F1 iz) + iNz,
sA = N r0 cos (F2 F1 iz) iNz,
(1)

(1)
vH = X F1 F2 F (1) (a) i ,

2(gs N )2 (1)
(1)
F1 F2 F (1) (a) i ,
X
gs N (2)
(2)
F F2 ,
wH = gs N r0 sin (F2 F1 iz) +
r0 sin 1
wA = gs N r0 sin (F2 F1 iz),
vA =
(D.23)
where
r02 =
3 2 (/2)
,
3 (/2)2 g2
2
.
6 (/2)2 2g2
(D.24)
This is nothing other than (4.33). Note that we have effectively replaced C by the parameter
to make the effective rotation of D4/D4 stacks illustrated in Fig. 11 manifest. The final B-period
constraint from (D.3) now fixes in terms of
4

v0

2i = r0 N L(a, ) + L(a, ) + ln i ,
(D.25)

where L(a, ) is defined in (3.51) and a = /2 (D.20).
As discussed in the text, we can consider this curve in the limit gs N 0 and arrive at an
approximate solution with boundary conditions (4.17)

s = r0 N (F1 F2 + iz) + c.c. + iN(z + z ),

(1)
(1)
v = X F1 F2 F (1) (a) i ,

w = C F1(2) F2(2) .
(D.26)
D.2. Connection with IIB
Now that we have an approximate curve (D.26) for gs N 1 which lifts to an exact solution
of (D.1) and (D.2), we can compare its moduli to those which extremize the type IIB potential
(4.8). The easiest way to do this is to write V directly as a function of a and and simply check
whether or not a and satisfying (D.10), (D.20), and (D.25) correspond to an extremum. This is
straightforward to do using the expression (3.52) for the period matrix of our hyperelliptic curve.
The general expressions for derivatives of V are somewhat complicated but if we set a = /2
359
and Re( ) = 0 they simplify greatly. For a V , we find in this case that
a V =
2N( + )(L
+ L i ) 4
aL
.
2
(L + L i )
(D.27)
Noting that
a L = 0
(D.28)
we must have + = 0. It is easy to see that (D.25) automatically has this property so indeed
a V = 0. One might be worried that we just derived a condition on boundary data, which is
supposed to be an initial condition for the curve, rather than the moduli. We only have a small
set of solutions in hand, though, corresponding to specific sorts of boundary conditions. Indeed,
we only obtained solutions that are invariant under (D.9). Implicit in this assumption is that we
choose boundary conditions consistent with this symmetry. This, in turn, requires + = 0.
We now turn our attention to V . This expression is also complicated but setting a = /2,
Re( ) = 0, and + = 0 we find that
V = 0
N (L + L i )

.
4
2i 1 2i
L
(D.29)
Noting that

1
g2
2
2
L =
(/2) (/2)
2i (/2)
3
(D.30)
we see that (D.29) is identical to (D.25) and hence the moduli of the reduced curve (D.26) are an
exact extremum of the IIB potential (4.8).
References
[1] K. Intriligator, N. Seiberg, D. Shih, Dynamical SUSY breaking in meta-stable vacua, JHEP 0604 (2006) 021, hepth/0602239.
[2] H. Ooguri, Y. Ookouchi, Landscape of supersymmetry breaking vacua in geometrically realized gauge theories,
Nucl. Phys. B 755 (2006) 239, hep-th/0606061.
[3] S. Forste, Gauging flavour in meta-stable SUSY breaking models, Phys. Lett. B 642 (2006) 142, hep-th/0608036.
[4] A. Amariti, L. Girardello, A. Mariotti, Non-supersymmetric meta-stable vacua in SU(N ) SQCD with adjoint matter,
JHEP 0612 (2006) 058, hep-th/0608063.
[5] D. Shih, Spontaneous R-symmetry breaking in ORaifeartaigh models, hep-th/0703196.
[6] K. Intriligator, N. Seiberg, D. Shih, Supersymmetry breaking, R-symmetry breaking and metastable vacua, hepth/0703281.
[7] H. Ooguri, Y. Ookouchi, C.S. Park, Metastable vacua in perturbed SeibergWitten theories, arXiv: 0704.3613 [hepth].
[8] S. Franco, A.M. Uranga, Dynamical SUSY breaking at meta-stable minima from D-branes at obstructed geometries,
JHEP 0606 (2006) 031, hep-th/0604136.
[9] I. Garcia-Etxebarria, F. Saad, A.M. Uranga, Local models of gauge mediated supersymmetry breaking in string
theory, JHEP 0608 (2006) 069, hep-th/0605166.
[10] H. Ooguri, Y. Ookouchi, Meta-stable supersymmetry breaking vacua on intersecting branes, Phys. Lett. B 641
(2006) 323, hep-th/0607183.
[11] S. Franco, I. Garcia-Etxebarria, A.M. Uranga, Non-supersymmetric meta-stable vacua from brane configurations,
JHEP 0701 (2007) 085, hep-th/0607218.
[12] I. Bena, E. Gorbatov, S. Hellerman, N. Seiberg, D. Shih, A note on (meta)stable brane configurations in MQCD,
JHEP 0611 (2006) 088, hep-th/0608157.
360
[13] C. Ahn, Brane configurations for nonsupersymmetric meta-stable vacua in SQCD with adjoint matter, Class. Quantum Grav. 24 (2007) 1359, hep-th/0608160.
[14] C. Ahn, M-theory lift of meta-stable brane configuration in symplectic and orthogonal gauge groups, Phys. Lett.
B 647 (2007) 493, hep-th/0610025.
[15] R. Argurio, M. Bertolini, S. Franco, S. Kachru, Gauge/gravity duality and meta-stable dynamical supersymmetry
breaking, JHEP 0701 (2007) 083, hep-th/0610212.
[16] M. Aganagic, C. Beem, J. Seo, C. Vafa, Geometrically induced metastability and holography, hep-th/0610249.
[17] R. Tatar, B. Wetenhall, Metastable vacua, geometrical engineering and MQCD transitions, JHEP 0702 (2007) 020,
hep-th/0611303.
[18] O. Lebedev, V. Lowen, Y. Mambrini, H.P. Nilles, M. Ratz, Metastable vacua in flux compactifications and their
phenomenology, JHEP 0702 (2007) 063, hep-ph/0612035.
[19] C. Ahn, Meta-stable brane configuration with orientifold 6 plane, hep-th/0701145.
[20] C. Ahn, More on meta-stable brane configuration, hep-th/0702038.
[21] J.J. Heckman, J. Seo, C. Vafa, Phase structure of a brane/anti-brane system at large N , hep-th/0702077.
[22] C. Ahn, Meta-stable brane configuration and gauged flavor symmetry, hep-th/0703015.
[23] A. Giveon, D. Kutasov, Gauge symmetry and supersymmetry breaking from intersecting branes, hep-th/0703135.
[24] R. Argurio, M. Bertolini, S. Franco, S. Kachru, Metastable vacua and D-branes at the conifold, hep-th/0703236.
[25] S. Murthy, On supersymmetry breaking in string theory from gauge theory in a throat, hep-th/0703237.
[26] S. Hirano, Models with quartic potential of dynamical SUSY breaking in meta-stable vacua, hep-th/0703272.
[27] C. Ahn, Meta-stable brane configuration of product gauge groups, arXiv: 0704.0121 [hep-th].
[28] I. Garcia-Etxebarria, F. Saad, A.M. Uranga, Supersymmetry breaking metastable vacua in runaway quiver gauge
theories, arXiv: 0704.0166 [hep-th].
[29] T. Kawano, H. Ooguri, Y. Ookouchi, Gauge mediation in string theory, arXiv: 0704.1085 [hep-th].
[30] C. Angelantonj, E. Dudas, Metastable string vacua, arXiv: 0704.2553 [hep-th].
[31] M.R. Douglas, J. Shelton, G. Torroba, Warping and supersymmetry breaking, arXiv: 0704.4001 [hep-th].
[32] C. Ahn, Meta-stable brane configurations with five NS5-branes, arXiv: 0705.0056 [hep-th].
[33] S.A. Abel, C.S. Chu, J. Jaeckel, V.V. Khoze, SUSY breaking by a metastable ground state: Why the early universe
preferred the non-supersymmetric vacuum, JHEP 0701 (2007) 089, hep-th/0610334.
[34] W. Fischler, V. Kaplunovsky, C. Krishnan, L. Mannelli, M.A.C. Torres, Meta-stable supersymmetry breaking in a
cooling universe, JHEP 0703 (2007) 107, hep-th/0611018.
[35] H. Abe, T. Higaki, T. Kobayashi, Y. Omura, Moduli stabilization, F-term uplifting and soft supersymmetry breaking
terms, Phys. Rev. D 75 (2007) 025019, hep-th/0611024.
[36] S.A. Abel, J. Jaeckel, V.V. Khoze, Why the early universe preferred the non-supersymmetric vacuum. II, JHEP 0701
(2007) 015, hep-th/0611130.
[37] M. Dine, J. Mason, Gauge mediation in metastable vacua, hep-ph/0611312.
[38] R. Kitano, H. Ooguri, Y. Ookouchi, Direct mediation of meta-stable supersymmetry breaking, Phys. Rev. D 75
(2007) 045022, hep-ph/0612139.
[39] C. Csaki, Y. Shirman, J. Terning, A simple model of low-scale direct gauge mediation, hep-ph/0612241.
[40] S.A. Abel, V.V. Khoze, Metastable SUSY breaking within the standard model, hep-ph/0701069.
[41] L. Anguelova, R. Ricci, S. Thomas, Metastable SUSY breaking and supergravity at finite temperature, hep-th/
0702168.
[42] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[43] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001) 3,
hep-th/0103067.
[44] F. Cachazo, S. Katz, C. Vafa, Geometric transitions and N = 1 quiver theories, hep-th/0108120.
[45] F. Cachazo, B. Fiol, K.A. Intriligator, S. Katz, C. Vafa, A geometric unification of dualities, Nucl. Phys. B 628
(2002) 3, hep-th/0110028.
[46] F. Cachazo, C. Vafa, N = 1 and N = 2 geometry from fluxes, hep-th/0206017.
[47] H. Ooguri, C. Vafa, Two-dimensional black hole and singularities of CY manifolds, Nucl. Phys. B 463 (1996) 55,
hep-th/9511164.
[48] D. Kutasov, Orbifolds and solitons, Phys. Lett. B 383 (1996) 48, hep-th/9512145.
[49] A. Karch, D. Lust, D.J. Smith, Equivalence of geometric engineering and HananyWitten via fractional branes,
Nucl. Phys. B 533 (1998) 348, hep-th/9803232.
[50] K. Dasgupta, K. Oh, R. Tatar, Geometric transition, large N dualities and MQCD dynamics, Nucl. Phys. B 610
(2001) 331, hep-th/0105066.
[51] K.h. Oh, R. Tatar, Duality and confinement in N = 1 supersymmetric theories from geometric transitions, Adv.
Theor. Math. Phys. 6 (2003) 141, hep-th/0112040.
361
[52] S. Mukhi, N.V. Suryanarayana, D. Tong, Braneantibrane constructions, JHEP 0003 (2000) 015, hep-th/0001066.
[53] S. Mukhi, N.V. Suryanarayana, A stable non-BPS configuration from intersecting branes and antibranes, JHEP 0006
(2000) 001, hep-th/0003219.
[54] E. Witten, Solutions of four-dimensional field theories via M-theory, Nucl. Phys. B 500 (1997) 3, hep-th/9703166.
[55] I. Affleck, M. Dine, N. Seiberg, Dynamical supersymmetry breaking in supersymmetric QCD, Nucl. Phys. B 241
(1984) 493.
[56] E. Witten, Branes and the dynamics of QCD, Nucl. Phys. B (Proc. Suppl.) 68 (1998) 216.
[57] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69, hep-th/9906070;
S. Gukov, C. Vafa, E. Witten, Nucl. Phys. B 608 (2001) 477, Erratum.
[58] A. Giveon, D. Kutasov, Brane dynamics and gauge theory, Rev. Mod. Phys. 71 (1999) 983, hep-th/9802067.
[59] K. Hori, H. Ooguri, Y. Oz, Strong coupling dynamics of four-dimensional N = 1 gauge theories from M theory
fivebrane, Adv. Theor. Math. Phys. 1 (1998) 1, hep-th/9706082.
[60] J. de Boer, Y. Oz, Monopole condensation and confining phase of N = 1 gauge theories via M-theory fivebrane,
Nucl. Phys. B 511 (1998) 155, hep-th/9708044.
[61] J. de Boer, S. de Haro, The off-shell M5-brane and non-perturbative gauge theory, Nucl. Phys. B 696 (2004) 174,
hep-th/0403035.
[62] T.H. Buscher, Path integral derivation of quantum duality in nonlinear sigma models, Phys. Lett. B 201 (1988) 466.
[63] A. Sen, KaluzaKlein dyons in string theory, Phys. Rev. Lett. 79 (1997) 1619, hep-th/9705212.
[64] A. Sen, Dynamics of multiple KaluzaKlein monopoles in M and string theory, Adv. Theor. Math. Phys. 1 (1998)
115, hep-th/9707042.
[65] D. Tong, NS5-branes, T-duality and worldsheet instantons, JHEP 0207 (2002) 013, hep-th/0204186.
[66] J.A. Harvey, S. Jensen, Worldsheet instanton corrections to the KaluzaKlein monopole, JHEP 0510 (2005) 028,
hep-th/0507204.
[67] A. Lawrence, J. McGreevy, Local string models of soft supersymmetry breaking, JHEP 0406 (2004) 007, hep-th/
0401034.
[68] I.A. Bandos, A. Nurmagambetov, D.P. Sorokin, The type IIA NS5-brane, Nucl. Phys. B 586 (2000) 315, hep-th/
0003169.
[69] R.A. Janik, Exact U (Nc ) U (N1 ) U (N2 ) factorization of SeibergWitten curves and N = 1 vacua, Phys. Rev.
D 69 (2004) 085010, hep-th/0311093.
[70] R. Dijkgraaf, C. Vafa, Matrix models, topological strings, and supersymmetric gauge theories, Nucl. Phys. B 644
(2002) 3, hep-th/0206255.
[71] R. Dijkgraaf, C. Vafa, On geometry and matrix models, Nucl. Phys. B 644 (2002) 21, hep-th/0207106.
[72] R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[73] J. de Boer, K. Hori, H. Ooguri, Y. Oz, Khler potential and higher derivative terms from M theory five-brane, Nucl.
Phys. B 518 (1998) 173, hep-th/9711143.
[74] J.H. Brodie, On mediating supersymmetry breaking in D-brane models, hep-th/0101115.
[75] D. Berenstein, Reverse geometric engineering of singularities, JHEP 0204 (2002) 052, hep-th/0201093.
[76] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
[77] J.F.G. Cascales, A.M. Uranga, M5-brane geometries, T-duality and fluxes, JHEP 0401 (2004) 021, hep-th/0307156.
Anomaly holography
Ben Gripaios a,b , Stephen M. West a,
a Rudolf Peierls Centre for Theoretical Physics, University of Oxford, 1 Keble Rd., Oxford OX1 3NP, UK
b Merton College, Oxford OX1 4JD, UK
Received 18 May 2007; received in revised form 25 July 2007; accepted 21 August 2007
Abstract
We consider, in the effective field theory context, anomalies of gauge field theories on a slice of a fivedimensional, anti-de Sitter geometry and their four-dimensional, holographic duals. A consistent effective
field theory description can always be found, notwithstanding the presence of the anomalies and without
modifying the degrees of freedom of the theory. If anomalies do not vanish, the d = 4 theory contains additional pseudoscalar states, which are either present in the low-energy theory as physical, light states, or are
eaten by (would-be massless) gauge bosons. We show that the pseudoscalars ensure that global anomalies
of the four-dimensional dual satisfy the t Hooft matching condition and comment on the relevance for
warped models of electroweak symmetry breaking.
1. Introduction
Maldacenas AdS/CFT correspondence [1], which has now been around for a decade or so,
has provided us with remarkable insights into strongly-coupled gauge theories, and has passed
numerous consistency checks. One such check, pointed out by E. Witten [2], is that Chern
Simons terms in the action on the AdS side, which are not gauge-invariant on the projective
boundary of AdS, correctly reproduce the anomaly structure of the global R-symmetry currents
on the CFT side.
It is believed that the correspondence is rather general and can, in particular, be extended to
cover the situation where the AdS space is truncated by one or more branes [35]. On the CFT
side, this is interpreted as a breaking of the conformal symmetry. If the theory on the AdS side
E-mail addresses: b.gripaios1@physics.ox.ac.uk (B. Gripaios), s.west1@physics.ox.ac.uk (S.M. West).

doi:10.1016/j.nuclphysb.2007.08.008
B. Gripaios, S.M. West / Nuclear Physics B 789 (2008) 362381
363
contains fermions, the presence of the branes leads to additional, anomalous contributions to
gauge transformations in the d = 5 theory. These contributions have already been discussed for
a flat extra-dimensional geometry, for example, in [612] and for a warped geometry in [13];
a clarification of these contributions and the rles they play in the effective field theories on both
sides of the AdS/CFT correspondence will be the goal of this paper. We will also discuss the
relevance for warped models of electroweak symmetry breaking.
We consider the correspondence in its crudest form, in which a gauge theory in a slice of
an AdS5 space (the so-called RS1 geometry [14]), weakly-coupled to itself and to gravity, is
dual1 to some gauge theory in four dimensions with large rank and strong coupling. This fourdimensional theory has a conformal invariance that is broken explicitly by an ultra-violet (UV)
cut-off and is non-linearly realized in the infra-red (IR).
We find that a consistent effective field theory (EFT) description can always be found, notwithstanding the presence of anomalies. In seeking this description, we take great care not to modify
the existing degree-of-freedom content of the theory. The resulting low-energy spectrum in d = 4
is not that which might be obtained by a nave dimensional reduction of the d = 5 theory. Indeed,
we find that a consistent description generically contains extra, scalar states.2 These are either
present in the low-energy effective theory as light, physical states, or are eaten by gauge bosons,
producing a theory with fewer massless gauge bosons in the low-energy spectrum. These extra
scalar states play an essential rle in the AdS/CFT correspondence: they lead to WessZumino
Witten (WZW) terms in the low-energy effective action that guarantee that global anomalies in
the d = 4 dual theory obey the t Hooft matching condition [15] at all energy scales.
Models of this type, in which a local symmetry group G in the bulk of AdS5 is broken to
subgroups H0 and H1 on the branes (located at positions z0 and z1 , respectively, in the fifthdimensional co-ordinate) have been extensively invoked as models of electroweak symmetry
breaking (EWSB), both with [16,17] and without [18] Higgs scalars. They provide natural models in which the hierarchy between the Planck and electroweak scales is explained either by the
warp-factor of the AdS geometry on the d = 5 side, or equivalently by the slow running of the
coupling constants on the d = 4 side. The effective theory at, or below, the weak scale is that
of an H0 -gauged G/H1 non-linear sigma model, with the massive gauge bosons corresponding
to generators in H0 , but not in H1 , integrated out. We show that specifying a plausible fermion
content for one such model, the so-called MCHM5 , renders it manifestly anomaly-free. Doing
the same for other models results in non-vanishing fermion anomalies, which will need to be
cancelled by adding ChernSimons terms or, more generally, altering the fermion content; such
changes, of course, affect the phenomenology, which we hope to explore in future work [19].
Although our findings apply to the general case of arbitrary bulk gauge group G broken to
subgroups H0,1 on the branes, most of the pertinent features are already extant in the simplest
case where G = U (1), and we shall use this as our primary example in the sequel.
The outline is as follows. In the next section we review the contributions to gauge-variance
in d = 5, coming from ChernSimons terms and fermionic anomalies, and their relevance in
the EFT context. In Section 3, we discuss the d = 4 holographic interpretation. In Section 4 we
discus the implications for models of EWSB.
The work presented here overlaps with a recent preprint [20], which, in particular, describes
the holographic connection between the d = 5 ChernSimons term and WZW terms in d = 4.
1 Up to energy scales of order of the AdS curvature scale, where the volume of the internal manifold, M, of the full
AdS M geometry, is resolved.
2 Actually, these are pseudoscalars but we do not make the distinction in what follows.
364
2. Gauge-variance in d = 5
Consider a gauge theory on a d = 5 spacetime that is topologically of the form R4 I ,
where I is a closed interval of the real line. We use co-ordinates x on R4 and z on the interval
I = [z0 , z1 ]. We refer to the disjoint boundaries of the interval as the z0 , or UV, brane and the z1 ,
or IR, brane. Equivalently [21], one can think of a space obtained as some orbifold of the space
R4 S 1 . We shall always use the interval formulation however, since it is more convenient for
dealing with the general boundary conditions (BCs) that we shall employ.
Imposing the requirement of gauge-invariance in the bulk does not necessarily imply gaugeinvariance on the branes, either at the classical or quantum level. Indeed, consider a Chern
Simons term in the action for the U (1) theory in the bulk of the form3

SCS = c d 4 x dz MNP QR AM FN P FQR .
(1)
Under a d = 5 gauge transformation of the form
AM AM + M (x, z),
we find
S = c

= 2c
(2)

d 4 x dz 5 F F (z z0 ) + (z z1 )

d 4 x dz F F (z z0 ) + (z z1 ) ,
(3)
where F = 12 F . So the action is not gauge-invariant on the branes even at the classical
level, in the presence of the ChernSimons term.
This is not necessarily the only source of gauge-variance however. Charged fermions, either
propagating in the bulk or localized on the branes, can also lead to gauge-variance at the quantum
level, via anomalies.
Consider first a bulk fermion of charge Q with respect to the U (1) gauge symmetry. The
smallest irreducible representation of the Dirac algebra in d = 5 has dimension four and is carried
by a Dirac spinor . In a non-compact theory, this precludes the existence of a perturbative gauge
anomaly, but this is not true in the presence of the branes, where different boundary conditions
for the two Weyl spinors making up make the theory intrinsically chiral, as is clear from the
existence of chiral zero modes in the d = 4 spectrum.
To see the resulting anomaly, let us take (x , z) as Poincar co-ordinates on a slice of AdS5
with curvature scale k such that the metric (signature mostly-plus) is given by
ds 2 =

1
dx dx + dz2 .
2
(kz)
The action for , including coupling to the gauge field, is

i
S = d 4 x dz g M DM M + H.c.,
2
(4)
(5)
3 We use Latin majuscules for d = 5 indices, e.g. M {0, 1, 2, 3, 5}, and Greek minuscules for d = 4, e.g.
{0, 1, 2, 3}. Our conventions for Dirac matrices, etc., are those of [22].
365
M A are curved-space gamma matrices and D =

where M is the bulk Dirac mass, M = eA
M
M + M iQAM includes the spin connection. In terms of the re-scaled Weyl spinors,
and , defined such that

2
= (kz)
(6)
,

the action is
S=
( iQA )
d x dz i ( iQA ) i
4
1
5 iQA5 )
(5 iQA5 ) + (5 + iQA5 ) (
2
M
(5 + iQA5 ) + ( + ) .
kz
+
(7)
Requiring that the variation of the action vanishes in the bulk gives the bulk equations of motion,
M
,
kz
M
0 = i ( iQA ) (z iQAz ) + .
kz
0 = i ( iQA ) (5 iQA5 ) +
Requiring that the variation of the action vanishes on the branes gives the condition

1
d 4 x [ + ]zz10 ,
0=
2
(8)
(9)
which is satisfied by BCs of the form [23]
= N ,
(10)
at z = z0 , z1 , where N = N (N may include boundary derivatives, and the signs account

for partial integrations of these).
Let us focus on just a subset of the possible BCs, where either or is set to zero on each
brane. We label the various possibilities by the ordered pair (0 , 1 ), where the first entry refers
to the z0 brane and the second entry to the z1 brane. The entries 0 and 1 take values in {+, };
a plus indicates that vanishes on the relevant brane, whereas a minus indicates that vanishes.
The reason for this notation will become evident when we consider the anomalies arising from
the bulk fermion with the specified boundary conditions.
Before we do that, let us point out that [21] the boundary conditions (+, +) give rise to a
left-handed massless Weyl fermion in d = 4, the (, ) conditions give rise to a right-handed
Weyl fermion, and the other possible BCs do not give rise to massless modes.
Returning to the anomalies, let us consider the effective action, [AM ], in d = 5 obtained by
integrating out the bulk fermion in a U (1) gauge field background, viz.

exp i [AM ] = D D exp iS[, AM ].
(11)
As a result of the anomaly, [AM ] is not invariant under a background gauge transformation of
the form (2). The variation for the BCs (0 , 1 ) is easily determined by comparison with results
366
previously obtained for S 1 /Z2 [7] and S 1 /Z2 Z2 orbifolds [8], and is given by

[AM ] = d 4 x dz (x, z)A(x, z),
(12)
where

Q3
F F 0 (z z0 ) + 1 (z z1 ) .
(13)
96 2
Let us pause to examine this result. Firstly, we see that the anomalies are localized on the branes.
This should come as no surprise, since it is only via the boundary conditions that any notion
of chirality is introduced. The anomaly is a topological artefact, and for the same reason, the
result is independent of the metric and the bulk mass M of the fermion. Secondly, the numerical
factor deserves comment. There is an extra factor of two in the denominator relative to the usual
consistent anomaly of a d = 4 Weyl fermion [24]. This extra factor of two indicates that the
anomaly is split between the two branes: if we integrate A(x, z) over the z co-ordinate, we find a
d = 4 anomaly whose value is the consistent anomaly of a left-handed Weyl fermion, multiplied
1
by a factor 0 +
2 , which takes values +1, 1, or 0 for the BCs (+, +), (, ), or (, )
respectively. We see that this d = 4 anomaly is in one-to-one correspondence with the anomaly
produced by the d = 4 massless fermion modes.
Anomalies can also occur due to brane-localized fermions. Unsurprisingly, they result in a
gauge-variation of the effective action (obtained in the usual way by integrating over the branelocalized fermions) that is localized on the relevant brane and whose value there is given by the
usual d = 4 anomaly. For example, a left-handed Weyl fermion of charge Q on the z0 brane
results in a variation
A(x, z) =
Q3
(14)
F F (z z0 ).
48 2
So there are three possible contributions to gauge-variance in theories of this type, viz. from
ChernSimons terms for bulk gauge fields, and from bulk and brane-localized fermions. All
contributions are localized on the branes and, in a convenient abuse of notation, we shall refer
to them collectively as brane-localized anomalies (though the classical variation of the Chern
Simons term is not a quantum anomaly).
None of these observations is new, and indeed the implications of the brane-localized anomalies for physics have already been discussed at length in the literature [612]. In the original
work [7] (formulated on a flat S 1 /Z2 orbifold), it was shown that the usual d = 4 anomaly
cancellation condition, applied to the d = 4 fermion zero modes, was a sufficient condition for
cancellations of the d = 5 brane-localized anomalies. But this is not true for a more general orbifold (as was pointed out for the orbifold S 1 /Z2 Z2 in [8]). Insufficiency is easily exhibited in
the interval formulation by means of a counter-example: consider a bulk fermion, charged under
a bulk U (1) gauge symmetry, with (+, ) BCs. Such BCs do not admit a chiral zero mode, so
there is no d = 4 anomaly. However, the d = 5 brane-localized anomalies are non-vanishing,
albeit equal and opposite.
In our counter-example (and in the example of [8]), the brane-localized anomalies can be
cancelled, without modifying the degrees of freedom of the theory, by a ChernSimons term
with an appropriate coefficient [9], rendering both the d = 4 and d = 5 theories anomaly-free.
But it is easy in the interval formulation to construct a different counter-example, where anomaly
cancellation cannot be achieved by ChernSimons terms: consider instead bulk gauge group
G = SU(2), broken on both branes to the U (1) subgroup generated by T3 . One possibility for the
A(x, z) =
367
d = 5 fermion content is to put a brane-localized left-handed Weyl fermions on each brane but
with opposite U (1) charge. There is no net anomaly in the spectrum of fermion zero-modes, yet
the equal and opposite anomalies on the branes cannot be cancelled by the non-Abelian Chern
Simons term,

3i
4
MNP QR
SCS = c d x dz
Tr AM N AP Q AR AM AN AP Q AR
2

3
AM AN AP AQ AR ,
(15)
5
which is proportional to Tr(T a {T b , T c }) and therefore vanishes for SU(2). Alternatively, we can
take the fermion content to be a bulk fermion transforming as a doublet under the bulk SU(2).
The BCs for the SU(2) doublet ( 1 2 )T need only respect the residual U (1) symmetry on the
branes; we take them to be (+, ) for 1 , which has charge +1 under the U (1), and (, +)
for 2 , which has charge 1. Computing the anomaly as above, we find

A(x, z) = 2 (z z0 ) (z z1 ) .
(16)
Now, there are no massless fermion modes in d = 4 and hence no d = 4 anomaly. But, yet
again, we have non-vanishing brane-localized anomalies in d = 5 that cannot be cancelled by a
ChernSimons term.
That ChernSimons terms do not suffice to cancel the brane-localized anomalies in d = 5,
given the d = 4 zero mode anomalies vanish, was already observed in [10] for orbifold theories
in arbitrary dimensions, with the ChernSimons terms generalized to four-form GreenSchwarz
[25] fields.
Finally, it was noted that U (1) anomalies of the type occurring in our SU(2) counter-example
could be cancelled by the addition of GreenSchwarz two-form bulk fields [26], or twisted
RamondRamond brane-localized fields [27]. But in adding such fields, it would appear, at least
navely, that the degree-of-freedom content of the theory is being changed.
Instead of trying to cancel the brane-localized anomalies in this way, we should like to follow
a different tack, motivated by the knowledge that we are dealing with theories in d = 5, that
are inherently non-renormalizable. They can, at best, be considered as EFTs, valid up to some
UV cut-off. In a renormalizable theory, the reasons for requiring anomaly cancellation are twofold: anomalies spoil both renormalizability and unitarity. In a non-renormalizable theory, as
Preskill [28] has pointed out, the only relevant issue is: what is the cut-off scale, beyond which
strong-coupling, unitarity violation, or other problems occur? He has, moreover, given lucid
arguments that show that one can always find a consistent EFT description of an anomalous
gauge theory, valid up to some non-vanishing cut-off, provided one allows the anomalous local
symmetries to be non-linearly realized. Thus we should expect that a consistent description of an
anomalous theory in d = 5 can always be found, irrespective of whether or not anomalies can be
made to cancel, and without changing the degrees of freedom of the theory. Our principal aim
will be to find this description.
What is more, there are two arguments that suggest that this approach will be instructive in
the context of the AdS/CFT correspondence. Firstly, the coefficient of the ChernSimons term is
a measure of the fermion content of the CFT [2]. We should like to consider an arbitrary CFT,
so we should also consider an arbitrary coefficient for the ChernSimons term, in which case the
brane-localized anomalies will, in general, be non-vanishing.
Secondly, the AdS/CFT correspondence tells us that the z co-ordinate in AdS5 corresponds,
roughly speaking, to the energy scale in the d = 4 dual. What is more, the branes at z0 and z1
368
somehow correspond to the UV and IR of the d = 4 dual, respectively. So the localization of the
anomalies on the branes in d = 5 should somehow encode information about the anomalies in
the UV and IR of the d = 4 dual. What we shall find is rather satisfying: the consistency of the
EFT description in d = 5 ensures that the global anomalies of the d = 4 dual satisfy the t Hooft
matching condition at all energy scales.
How, then, do we find a consistent EFT description of a theory in d = 5 with brane-localized
gauge anomalies, without changing the degrees of freedom of the theory? The problem with
anomalies is that they lead to a loss of gauge-invariance. In a gauge-invariant theory, on the other
hand, constructing a consistent EFT description is straightforward, because gauge invariance
furnishes us with a set of equivalent descriptions of the theory, any of which can be invoked
as ones whim dictates. Then, for example, the strong-coupling scale (at which calculability is
lost) is easily determined by power counting in a gauge in which the propagator is well-behaved,
e.g. the t HooftFeynman gauge. Then it is easy to see that the strong-coupling scale is the true
cut-off, because Lorentz invariance and unitarity are easily exhibited up to the strong-coupling
scale, by working in a covariant gauge or unitary gauge, respectively.
But in an anomalous theory, the gauge-invariance, and its associated benefits, seem to be
absent. All is not lost, however, once we realize that gauge symmetry is not really a symmetry
at all, but rather (in the sense discussed above) a redundancy, a set of equivalent descriptions.
It is, furthermore, a redundancy that is easily resurrected, by adding dynamical scalar fields
transforming under the gauge group, and including terms involving the scalar fields that cancel
the original anomalous variations coming from the fermions. To show that the new description
with additional scalar fields and non-anomalous gauge symmetry is equivalent to the old one, it
suffices (at least locally) to choose the gauge in which the scalars vanish. We are then left with
a theory without the scalars and without the gauge symmetry, viz. the original description. We
stress that adding the scalars in this way does not change the degree-of-freedom content of the
theory.
Following Preskill [28], let us see how this works in the case of a U (1) gauge theory in d = 4,
coupled to a left-handed Weyl fermion of charge Q. Under a background gauge transformation
A A + (x), the effective action obtained by integrating out the fermion varies, analogously to (12) and (13), as

[A ] =
d 4x
Q3
F F .
96 2
(17)
To resurrect the gauge invariance, introduce a scalar field , transforming as + under

the U (1), together with a non-renormalizable term in the Lagrangian of the form
L
Q3
F F .
96 2
(18)
It is then apparent that gauge invariance of the effective action is restored, and further that the
original description is recovered by the gauge choice = 0. This is not the end of the story,
however; in the spirit of EFT, we should include all terms in the effective Lagrangian consistent
with the symmetry, since they will be generated by quantum effects, even if we do not include
them a priori. In particular, kinetic terms for will be induced, at leading order in the derivative
expansion. Thus, the full effective Lagrangian takes the form
L=
369
1
f2
( A )2
F F i ( iQA ) +
2
2
4g
Q3
F F + ,
(19)
96 2
where the ellipsis denotes terms suppressed by powers of the scale f . Because of the gauge invariance, the theory is manifestly Lorentz-invariant and unitary, up to the cut-off scale at which
strong-coupling occurs, which we estimate on the basis of nave dimensional analysis to be
around 4f . To go back to the original description, we set = 0; we see that we have an effective theory of a gauge boson of mass gf , coupled to a massless Weyl fermion, and valid up
to a cut-off 4f . In the gauge-invariant description, the massive gauge boson arises because the
gauge symmetry is non-linearly realized by the Goldstone boson .
Having seen how the resurrection of gauge invariance is used to construct a consistent EFT
in d = 4, let us apply the same idea to construct a consistent EFT of an anomalous gauge theory
on an interval in d = 5. Consider again a U (1) gauge symmetry in the bulk, unbroken on the
branes at the classical quadratic level, but with gauge-variant contributions from arbitrary branelocalized anomalies of the form

Q3
F F 0 (z z0 ) + 1 (z z1 ) .
[AM ] = d 4 x dz
(20)
2
192
The theory is gauge-invariant in the bulk, but not, in general, on the branes. To make a manifestlyconsistent EFT, we resurrect the gauge symmetry on the branes by adding brane-localized scalars,
0 and 1 , transforming as
0 0 + (x, z0 ),
1 1 + (x, z1 ),
together with brane-localized interaction terms of the form

Q3
0 0 (z z0 ) + 1 1 (z z1 ) .
S d 4 x dz
F
F
192 2
(21)
(22)
Yet again, the spirit of EFT demands that we write down all terms consistent with the bulk U (1)
and other symmetries. Up to quadratic order in the derivative and field expansion (and disregarding the fermions), we have
S = Sbulk + S0 + S1 ,
where
1
1
MN
Sbulk = d x dz
2 FMN F
+ ,
(kz)5
4g

f2
1
S0 = d 4 x dz (z z0 ) 2 F F + 0 ( 0 A )2 + ,
2
4g0

f2
1
S1 = d 4 x dz (z z1 ) 2 F F + 1 ( 1 A )2 + .
2
4g1
(23)
(24)
Now the scales f0 and f1 are given by the natural scales on the respective branes, viz. 1/z0 and
1/z1 , and g0 and g1 are the couplings of brane-localized gauge kinetic terms.
370
We note that, at the quadratic level (which determines the spectrum), the only effect of nonvanishing brane-localized anomalies is to force the inclusion of brane-localized scalars in order
to maintain a fully gauge-invariant description. Now, the effect of adding brane-localized scalars
is well known [21]; it changes the spectrum of zero modes in d = 4, when we do the Kaluza
Klein expansion. To see this, we first need to fix the gauge. Let us define the -gauge by adding
to the bulk action the gauge-fixing term

2

1
A5
GF
Sbulk
(25)
= d 4 x dz
A
+
z
.
5
z
2 kzg 2
The gauge is not fixed completely. There are residual gauge transformations (x, z) such that
5

= 0,
+ z5
(26)
z
under which (25) is invariant. This is a second-order differential equation in z, whose solution
contains two arbitrary functions of x. To fix these two, residual, d = 4, gauge symmetries, we
also define the 0 - and 1 -gauges that correspond to adding the brane-localized terms,

2

1
A5
GF
4
2
A + 0 f0 0 + 2
S0 = d x dz (z z0 )
(27)
20
zg k
and

S1GF =
d 4 x dz (z z1 )

2
1
A5
,
A + 1 f12 1 2
21
zg k
(28)
respectively. These suffice to fix the gauge completely. Requiring that the variation of the action
vanishes in the bulk gives the bulk equations of motion for the A and A5 zero modes,

z5 z1 5 A = 0,
(29)

A5
5 z5
(30)
= 0.
z
The solutions of the bulk equations of motion are,
A = B + z 2 C ,
(31)
A5 = Dz log z + Ez,
(32)
B,
C,
D and E do not depend on z.

where
Requiring that the variation of the action vanishes on the branes gives the conditions,

1
2
f0,1 A 2 5 A A = 0,
g zk
z0,1

A5
2
A5 = 0,
z 5 (A5 /z) 0,1 f0,1
0,1 2
zg k
z0,1

A
5
2
f0,1
0,1 2 0,1 = 0,
zg k
z0,1
where the + and are for z = z0 and z = z1 , respectively.
(33)
(34)
(35)
371
Substituting the solutions for A and A5 into the boundary variations we find that the zero
modes are
A5 = Ez,
0 =
E
f02 g 2 k
and 1 =
E
f12 g 2 k
(36)
where E is unconstrained until fixed by the normalization of the zero mode. We see that we have
a scalar zero mode which is partly A5 , partly 0 and partly 1 . There is no vector zero mode.
Needless to say, the spectrum of zero modes is a physical and gauge-independent quantity,
though its description in terms of fields is not. Let us, then, reassure ourselves that computing
the spectrum in another gauge will give the same result. Most interesting among these is a gauge
in which the boundary scalars 0,1 vanish, because in this gauge we recover the original description of the anomalous gauge theory. Now the gauge-fixed action, S = Sbulk + S0 + S1 , has
contributions

1
1
A5
MN
Sbulk = d 4 x dz
F
F
A
+
z
+
MN
5
z
4(kz)5 g 2
2 kzg 2

f2
1
S0 = d 4 x dz (z z0 ) 2 F F + 0 A A + ,
2
4g0

f2
1
S1 = d 4 x dz (z z1 ) 2 F F + 1 A A + .
(37)
2
4g1
The bulk equations of motion for A and A5 are identical to the previous case and so the solutions
are those given in (31) and (32). The boundary variations for A are identical to those shown
in (33) and consequently there is no zero mode for A . The boundary variations for A5 are now
given by

5 (A5 /z)A5 z = 0.
(38)
0,1
Substituting the bulk solution for A5 into the above we find the zero mode
A5 = Ez,
(39)
where E is unconstrained. So again we find a scalar zero mode and no zero mode for A .
Yet another gauge is A5 = 0. In this gauge, the residual gauge transformations, (x, z) such
that 5 (x, z) = 0, contain only one arbitrary function of x, so the gauge is fixed completely by
a further gauge fixing on just one brane, say at z0 . So now the gauge-fixed action, S = Sbulk +
S0 + S1 , can be chosen to consist of

1
1
1
Sbulk = d 4 x dz
F
A
+
,
F
5 5
4
2
kzg 2

f2
1
4
S0 = d x dz (z z0 ) 2 F F + 0 ( 0 A )2
2
4g0

1
2
A + 0 f02 0 + ,
20

f12
1
4
2
( 1 A ) + .
S1 = d x dz (z z1 ) 2 F F +
(40)
2
4g1
372
In this gauge, it is useful to split the gauge field into transverse and longitudinal components
defined in d = 4 momentum space by

p p
AT =
A ,
p2
p p
AL
(41)
A .
=
p2
The bulk equations of motion for the zero modes are

z5 z1 5 AT,L = 0,
(42)
with solutions
AT = B + z 2 C ,
(43)
=D +z E .
(44)
The boundary variations in momentum space in this gauge are given by

1
2
f0,1
AT 2 5 AT AT = 0,
g zk
z
0,1

1

f02 AL + 2 5 AL AL
= 0,
g zk
z0

1

f12 AL + 2 5 AL f12 p 1 AL
= 0,
g zk
z1

0 0 z = 0,
0

L
p A 1 z = 0.
1
(45)
(46)
(47)
(48)
(49)
Substituting the bulk solutions for AT and AL into the above, we find that there is no zero
mode for AT , but there is a zero mode which is part AL and part 1 . The AL part is, however,
unphysical since it does not couple to conserved currents. We are thus left with the scalar zero
mode, 1 .
We see, exhaustively, that the spectrum of zero modes is the same in a variety of gauges, and,
in particular, in the gauge in which the original description of the anomalous theory is recovered.
The important point in this gauge is that a consistent EFT treatment requires that we include mass
terms for the gauge fields on the boundary (cf. (37)). In retrospect, this hardly seems surprising,
given that the gauge symmetry that would have forbidden such terms on the branes is anomalous.
(If we had chosen, perversely, not to include the mass terms at tree-level, they would be generated
nevertheless by loop effects.)
Let us compare with the spectrum of massless modes we would have obtained with vanishing
anomalies on both branes, i.e. with 0 = 1 = 0. Now gauge invariance is achieved without the
boundary scalars. The simplest way to find the spectrum is to choose A5 = 0 gauge. The action
in this case is

1
1
1
A
+
S = d 4 x dz
5 5
4
2
kzg 2

1
1
d 4 x dz (z z0 ) 2 F F + (z z1 ) 2 F F .
(50)
4g0
4g1
The bulk equation of motion and boundary variations are

z5 z1 5 A = 0,
and

5 A A z
0 ,z1
= 0,
373
(51)
(52)
respectively. We see that now there is a massless gauge boson in the spectrum given by A = B ,
where B is an undetermined constant in z. Similarly, we may consider the case where just one of
the brane-localized anomalies is non-vanishing. The spectrum contains neither vector nor scalar
zero modes.
Let us now compare our findings with those obtained previously in the string theory context.
By being careful not to change the degrees of freedom content of the theory, we have found that
the zero-mode spectrum depends solely on whether or not the brane-localized anomaly coefficients, 0 and 1 , are vanishing. It does not depend on their magnitude or relative sign. This is
clear from the discussion surrounding Eqs. (20)(24): 0 and 1 appear in the WZW interaction
terms, but not in the quadratic action; the latter determines the spectrum. To recap, if both 0
and 1 vanish, we get a vector zero mode and if both do not vanish, we get a scalar zero mode.
Otherwise, there are no zero modes.
In the previous literature, it was found (in the case that both anomalies are non-vanishing) that
cancelling anomalies by brane-localized twisted RamondRamond states [27] gave precisely this
pattern, whereas cancelling them by bulk GreenSchwarz two-forms [26] gave no zero mode for
0 = 1 , and both vector and scalar zero modes otherwise. Our explanation for this is as follows. Adding brane-localized states resurrects gauge-invariance on the branes without changing
the degree of freedom content of the theory (the brane-localized states can be gauged away).
Adding bulk two-forms, however, resurrects gauge-invariance on the branes, but also changes
the degrees of freedom of the theory. Indeed, it is clearly impossible to gauge away a bulk field
with the resurrected brane-localized gauge symmetry. In a sense then, our results are just an
effective field theorists vulgarization of the string theory results.
Let us now make two immediate generalizations. Firstly, we consider the case where the gauge
symmetry is assumed broken (or rather, non-linearly realized) ab initio on one or both branes.
The easiest way to realize this in the interval approach is to include a boundary scalar or scalars
by hand [21], just as one does for an anomaly. Of course, now there is no term of the form (18),
but as we have seen, this is irrelevant as far as determining the spectrum is concerned: it is the
mere presence of the scalar (forced upon us in the anomalous case) that changes the spectrum.
Secondly, we can consider what happens in the case of a general non-Abelian bulk gauge
group G broken to subgroups H0,1 on the respective branes. As in the last paragraph, this is
easily achieved in the interval approach via boundary scalars. Yet again, if any of the generators of the subgroups H0,1 has a brane-localized anomaly, we should add a boundary scalar for
that generator on the relevant brane, together with a non-Abelian WessZuminoWitten (WZW)
term [29], generalizing (18), to cancel the anomaly. The type of zero mode, if any, corresponding
to that generator is then determined by exactly the same considerations as in the U (1) case above.
So, for example, if a would-be massless gauge boson (corresponding to a generator in H0 H1 )
has an anomaly on one or both branes, we will find in its stead either no massless state, or a light
scalar.
We remark that, although such scalars appear massless at tree-level, they will acquire masses
via quantum loops of propagators stretching from one brane to the other. The masses arise because one brane does not respect the symmetry group of the other. Thus, boundary scalars, which
374
are Goldstone bosons of the subgroup on the relevant brane, are really pseudo-Goldstone bosons.
The resulting masses, which are non-local in origin, are finite, and are of order 1/z1 in magnitude, further suppressed by a loop factor of order 4 [16]. It is for this reason that we refer to
them as light scalars, rather than massless scalars.
Finally, we remind the reader that, although we have focused our attention on theories on a
warped interval (which admit a holographic dual), the anomaly considerations discussed here apply equally to a theory on an interval with arbitrary geometry. This is because the brane-localized
anomaly structure depends only on the topology of the interval and the associated fermionic
boundary conditions. If the brane-localized anomalies are non-vanishing, the construction of a
consistent EFT is easily done, by adding boundary scalars in the same fashion, and the same conclusion applies: the spectrum of d = 4 zero modes in the presence of a brane-localized anomaly
is altered, in that there are either extra, light scalars, or fewer massless gauge bosons.
3. Anomalies in d = 4
We saw in the preceding section how a consistent EFT description for an anomalous gauge
theory on an interval in d = 5, achieved by resurrecting gauge invariance everywhere on the
interval, implies extra scalar states in the theory. These scalar states are either present as light,
physical states in the low-energy spectrum in d = 4, or are eaten by would-be massless gauge
bosons, removing the gauge bosons from the low-energy spectrum in d = 4.
In this section, we explain the rle these scalar states play in the context of the AdS/CFT
correspondence for a warped geometry: they ensure t Hooft matching of anomalies of global
symmetries in the d = 4 dual, at all energy scales.
Before seeing how this comes about, we remind the reader of t Hoofts argument [15]. Consider a theory with some global symmetry group G that has an anomaly at some energy scale,
meaning that some correlation function of three global currents has non-vanishing divergence, or
equivalently that the divergence of the current is non-vanishing in the presence of a background
gauge field. (In typical examples, the theory is taken to be weakly-coupled at the given scale such
that the anomaly is calculable.) Now add spectator fermions, transforming in representations of G
so as to cancel the global G anomalies, and weakly-gauge the symmetry G, with gauge coupling
strength g
1. As we run down to a lower energy scale, the gauge theory we have constructed
remains, of course, anomaly-free. If we then further take the limit in which g 0, decoupling
the gauge fields, we again end up with a theory with anomaly-free global symmetry group G.
This theory still contains the decoupled spectator fermions (since they were only ever weakly
coupled to other sectors), whose contribution to the global G anomalies is the same as it was at
the higher scale. This implies that the global G anomalies of the original theory (without spectator fermions or gauge fields) cannot have changed either. That is, the global anomalies of the
original theory must match at all energy scales. This holds true even though the theory may have
gone through one or more strong-coupling transitions such that the effective weakly-coupled degrees of freedom (if any) may be completely changed. The weakly-coupled degrees of freedom
that contribute to the anomaly include fermions transforming in representations of G, together
with scalar fields, which can contribute to the anomaly via WessZuminoWitten (WZW) terms
in the effective action [29].
The d = 4 holographic dual of an AdS5 theory seems to lend itself ideally to a study of
t Hooft matching, in that the z co-ordinate (more precisely, its logarithm) corresponds to the
energy scale in the d = 4 CFT. This is because the combination of a constant scale transformation
x ax in the co-ordinates of the d = 4 dual, accompanied by a scaling z az, amounts to
375
an isometry of the AdS5 metric (4), and so does not change the physics. This is what we expect
for a theory with conformal symmetry, provided we interpret log z as the energy scale. Thus we
expect, intuitively, that the non-trivial z dependence of the anomalies in the d = 5 theory encodes
information about the anomalies at different energy scales in the d = 4 theory, which is precisely
the context in which the t Hooft anomaly matching condition is applied.
To set the scene, let us review Wittens original observation [2], relating the anomalies in the
d = 4 theory and the ChernSimons term in an AdS space without branes. We start from Wittens
conjecture for the AdS/CFT correspondence in the form

exp d 4 x O
(53)
=
D exp iS[],
CFT
(z0)
where the right-hand side represents a path-integral in AdS with respect to generic bulk field
(x, z), with the restriction that (x, z) tends to the value (x) on the projective boundary of
AdS, given by z 0. According to the conjecture, this is equivalent to the correlation function of
a CFT deformed by O, where O is some operator of the CFT dual to the source (x). In the
limit in which we are interested, where the CFT has large rank and strong (t Hooft) coupling,
the path integral on the right-hand side is understood to be computed on-shell, i.e. the action
is evaluated subject to the condition that the classical equations of motion are satisfied. As it
stands, the correspondence is ill-defined: the action on the right-hand side has a divergence that
comes from integrating the Lagrangian density over the IR of AdS (the region of small z). If the
correspondence has any chance of being true, it must be that the left-hand side is also divergent,
and indeed it is: the divergence is a UV divergence arising because we have deformed the CFT.
This corroborates our previous claim that log z corresponds to the energy scale in the d = 4 dual:
the region of small z in AdS5 corresponds to the UV region of the CFT. More specifically, we
can remove the divergence on the AdS side, by truncating AdS5 at some small value z = z0 ; this
truncation must correspond to cutting-off the deformed CFT in the UV in some definite fashion
at a scale 1/z0 . The conjectured correspondence is thus modified to

=
D exp iS[],
exp d 4 x O
(54)

CFT
(x,z0 )=(x)
where we have put a prime on the left-hand side to indicate that the deformed CFT has been
cut-off in the UV. We note in passing, that cutting-off the CFT in the UV will, in general, induce
kinetic terms for the generic source field (x) (a source is, from the EFT point of view, just a
higher-dimensional field), so we are free to make (x) a dynamical field if we so choose, by
performing a path-integral with respect to it on both sides of (54).
In the particular case of a bulk gauge field, AM (x, z), we might write the correspondence as

4
exp d x J a
(55)
=
DAM exp iS[AM ].

CFT
A (x,z0 )=a (x)

A5 (x,z0 )=0
For the time being, we choose not to path-integrate with respect to the boundary field a (x) =
A (x, z0 ), meaning this is a background gauge field in the d = 4 dual. Now, assuming that
the d = 5 integrand is gauge-invariant, we can show (at least formally) that the global CFT
current J is conserved. Indeed, consider making a change of variables in the d = 5 theory that
takes the form of a bulk gauge transformation, AM AM + M (x, z), with the restriction that
376
5 (x, z0 ) = 0. Since the integrand is gauge-invariant, we find

DAM exp iS[AM ] =
DAM exp iS[AM ],
A (x,z0 )=a (x)
A5 (x,z0 )=0
(56)
A (x,z0 )=a (x)+ (x)

A5 (x,z0 )=0
where (x, z0 ) = (x), implying

= 0.
exp d 4 x J a +
CFT
(x)
(57)
Thus

J e
d 4 x J a
CFT
= 0.
(58)
Going further, we may ask what happens if the d = 5 action contains a ChernSimons term such
that the action is gauge-invariant everywhere, except on the z0 boundary. For a ChernSimons
term of the form (1), the same argument implies that

4

J + cf f e d x J a CFT = 0,
(59)
where f = a a . So we find that the bulk ChernSimons term produces an anomaly in
the global current of the cut-off, deformed CFT, in the presence of the background gauge field a .
In particular, in the case of type IIB supergravity on AdS5 , one finds that the ChernSimons term
for SU(4) bulk gauge fields reproduces the global anomaly of the SU(4) R-symmetry currents in
the d = 4 dual, which is N = 4 supersymmetric YangMills theory [2,30].
On reflection, the argument we have just reviewed is suspect, because the path integral on the
right-hand side of (55) involves integration over infinitely many, physically-equivalent, gauge
field configurations. In order to make sense of this path-integral, we need to regulate it by, say, the
procedure of Faddeev and Popov, or more generally, BRST. But such a gauge-fixing procedure
invalidates the argument just given, which invoked gauge-invariance of the integrand in the d = 5
theory.
We can recover the argument in a refined form by regulating the d = 5 path integral in such
a way that it is rendered finite (at least at tree-level), but such that there remains a residual
bulk gauge symmetry to which a background gauge transformation (x) can be smoothly lifted.
A suitable regulator is provided by the delta-functional (A5 (x, z)) (for which the Faddeev
Popov determinant is trivial), whose argument is invariant under the residual gauge transformations (x, z) such that 5 (x, z) = 0, i.e. those for which (x, z) is independent of z. We
note that this is compatible with the boundary condition, A5 (x, z0 ) = 0, chosen on the z0 brane.
Furthermore, with this BC, the regulator (A5 (x, z)) can be used even when the theory is not
gauge-invariant on the z0 brane, because of an anomaly. This is because we have chosen A5 = 0
as a boundary condition there, for which the argument of the delta-functional vanishes identically. With the delta-functional regulator included, the correspondence becomes

4
exp d x J a
(60)
=
DAM A5 (x, z) exp iS[AM ],

CFT
A (x,z0 )=a (x)

A5 (x,z0 )=0
and the path integral on the right-hand side is rendered finite in such a way that the argument
given previously still goes through. We shall make frequent use of this argument, which we call
377
the holographic anomaly argument, in the sequel. We repeat that it can be employed whenever
the d = 5 theory is gauge-invariant for z > z0 .
Hitherto, we have only considered the possibility of gauge-variance on the z0 brane resulting
from ChernSimons terms. If we have a bulk fermion charged under the gauge group, then it
should also be present in the path integral on the right-hand side, including the path-integration
with respect to its value on the z0 -brane. With the types of fermionic BCs we have considered,
only one Weyl component of , say, can be non-vanishing and (60) is modified to

4

DD eiS0 ,a e d x J a +O CFT

iS0 ,a
= DD e
DAM D D eiS[AM , ] ,
A (x,z0 )=a (x)
A5 (x,z0 )=0
(x,z0 )=(x)
where we included a possible brane-localized action, S0 , for the fermion. When we invoke the
holographic anomaly argument, we will pick up an extra contribution to gauge-variance on the
z0 brane coming from the Jacobean of the bulk fermion measure under the gauge transformation.
This Jacobean is determined directly from (13), where, because there is currently only a single
brane (at z = z0 ), we discard the z1 piece. The holographic anomaly argument then gives
4

J + cf f e d x J a CFT = 0,
(61)
where now c contains contributions from both the ChernSimons terms and the bulk fermion.
Similarly, for a dynamical brane-localized fermion at z = z0 , we should take (60), include
the brane-localized action for the fermion, and path-integrate with respect to it, obtaining

4

D D eiS0 ,a e d x J a CFT =
DAM D D eiS0 [ ,a ] eiS[AM ] .
A (x,z0 )=a (x)
A5 (x,z0 )=0
(62)
Now when we make the holographic anomaly argument, we find an expression like (61), but now
the coefficient c contains contributions from ChernSimons terms and brane-localized fermions.
In this way, we see how the z0 -brane-localized gauge anomaly of the d = 5 theory computes,
via the holographic anomaly argument, the global anomaly of the d = 4 dual, which is a stronglycoupled CFT coupled to external fields, in a background gauge field. This anomaly is, of course,
the anomaly computed at the scale 1/z0 .
To see how t Hooft matching works in the d = 4 dual, we need to compute the anomaly at
some lower energy scale, 1/z , say. To do so, we need to compute the EFT for the d = 4 dual
obtained by integrating out the physics at energy scales corresponding to z0 < z < z . In the d = 5
picture, we do this by shifting the position of the z0 brane to z = z , whilst demanding that the
physics at z > z remains the same. At tree-level, this amounts to modifying the brane-localized
action at z such that the solutions of the d = 5 equations of motion remain the same for z > z .
The details of this holographic renormalization group flow are described in [31,32]; as far as
the anomalies are concerned though, the flow is trivial. The theory with brane at z contains the
same field content as the theory with brane at z0 , and therefore the brane-localized anomalies
remain the same. As a corollary, the global anomalies of the d = 4 dual cut-off at 1/z are the
same for all z > z0 , and t Hooft matching is trivial.
378
So far, we have explicitly excluded the brane at z = z1 from our arguments. Re-instating
the brane corresponds [3,4] to spontaneously breaking the conformal symmetry in the IR at
energy scale 1/z1 . At energy scales lower than 1/z1 , the correspondence no longer holds; the
d = 4 theory must be studied in the normal way, without recourse to holographic arguments.
We know from the previous section that adding the brane at z = z1 introduces further gauge
anomalies into the d = 5 theory. What, then, do these correspond to on the dual d = 4 side? In
trying to generalize the arguments we have just made for the theory with only a single brane
at z = z0 , we immediately encounter a problem: the presence of a non-vanishing anomaly on
the z1 brane prohibits us a priori from making the holographic anomaly argument, because this
assumed gauge invariance for z > z0 ; the existence of the anomaly implies that there is no gauge
symmetry at z = z1 . Unsurprisingly, the solution is to resurrect the gauge symmetry at z = z1 ,
by adding a scalar field 1 together with a brane-localized term of the form (18), just as we did
before. We stress again that we are not changing the theory or its degrees of freedom, but merely
furnishing ourselves with an equivalent description.
The beauty of this equivalent description, with resurrected gauge symmetry on the z1 brane,
is that we can invoke the holographic anomaly argument once more.4 When we do so, we find
that the global anomaly of the d = 4 dual at the scale 1/z0 is computed by the gauge anomaly on
the z0 brane in the d = 5 theory, and that by shifting the position of the z0 brane to z0 z < z1
whilst keeping the physics the same, the global anomaly trivially obeys the t Hooft matching
condition at all scales 1/z0 1/z > 1/z1 . What happens at z = z1 ? In the d = 5 picture, this
corresponds to the endpoint of the holographic RG flow, with the cut-off brane at z hitting the IR
brane at z1 . Now the entire bulk has been integrated out, and we are left with a d = 4 EFT with
cut-off 1/z1 , obtained by adding the z -brane-localized action (with z z1 ) to the z1 -branelocalized action [32]. If we make the holographic anomaly argument at z = z1 , we will find that
the global anomaly of the d = 4 dual cut-off at z1 is still the same as the anomaly at z < z1 :
contributions to the z1 -brane-localized anomaly, whether they come from ChernSimons terms,
or bulk or brane-localized fermions, are cancelled by the boundary scalar 1 , which appears as a
light state in the low-energy d = 4 theory. What happens at energy scales below 1/z1 ? Now the
correspondence breaks down, but we can still compute the effective action by taking the effective
theory with cut-off 1/z1 and integrating down to the lower energy scale in the normal way. The
global anomalies must still match, and, provided the degrees of freedom remain weakly-coupled,
come from the light fermions and scalars.
We remind the reader that, so far in this section we have kept the gauge field a (x) as a
non-dynamical, background gauge field, since this is what we need for computing the global
anomalies of the d = 4 dual theory. We can also choose to make the gauge field dynamical, by
path-integrating with respect to it on both sides of, e.g. (60). The fate of a light scalar (assuming
there is a non-vanishing anomaly for some generator on the z1 brane) then depends on whether
or not there is an anomaly for the generator on the z0 brane. If there is an anomaly on the z0
brane, then a mass term will be generated for the gauge field in the d = 4 dual EFT at the scale
1/z0 ; it will not be present in the low-energy spectrum, but the scalar will. If, by contrast, there
is no anomaly on the z0 brane, the dynamical gauge field will be present as a massless gauge
field in the theory cut-off at energy scales above 1/z1 . At the energy scale 1/z1 , it will eat the
4 Note that we have not bothered to resurrect the gauge symmetry on the z brane, since the holographic anomaly
0
argument is not contingent upon it.
379
scalar 1 , leaving neither a gauge boson nor a scalar in the low-energy spectrum. These spectra
are, of course, just those we determined in the previous section.
What would happen if we chose not to resurrect the gauge symmetry on the z1 brane by
adding the scalar 1 ? Now we cannot use the holographic anomaly argument to compute the
global anomalies of the d = 4 dual, because the theory is not gauge invariant for all z > z0 . That
said, the spectrum of d = 4 zero modes, including possibly a light scalar degree of freedom,
cannot change, since the spectrum cannot depend on which equivalent description of the theory
we choose. Reassuringly, the spectrum does not change: as we saw in the previous section, the
scalar is provided in this description by a zero mode of the A5 field, rather than by 1 . What
does change is that we cannot compute the global anomaly of the d = 4 dual in this description,
because we simply do not have a theorem, in the form of the holographic anomaly argument,
available to us.
4. Implications for warped models of EWSB
We have tacitly assumed a U (1) bulk gauge symmetry in the foregoing, but generalization
to the case where bulk gauge group G is broken to subgroups H0,1 on the respective branes is
straightforward. We simply analyse each generator of the bulk Lie algebra in turn. If the generator
is assumed to be unbroken on a given brane (i.e. contained in H0 or H1 , as appropriate) then we
should consider whether there is an anomaly, involving that generator, localized on the given
brane, due to any of the sources we discussed above. If there is an anomaly, then we can find a
consistent EFT by adding a boundary scalar to restore gauge invariance. This means, though, that
just as in the U (1) case, the spectrum of massless states in d = 4 will be changed, with either
extra light scalars or fewer massless gauge bosons. So, in general, all of the H0,1 , anomalies
on the respective branes must be cancelled in order that the spectrum of massless states be that
which is assumed a priori.
This observation is of some relevance for warped models of electroweak symmetry-breaking
in the Standard Model (SM).
Take, as an example, the Composite Higgs Model of [17]. This model employs a bulk gauge
group G = SU(3)c SO(5) U (1)X , which is broken down to H1 = SU(3)c O(4) U (1)X on
the IR boundary, and H0 = SU(3)c SU(2)L U (1)Y on the UV boundary, where hypercharge
is defined as Y = X + T3R . The Higgs appears as a pseudo-Goldstone Boson consisting of the
four real scalar fields that corresponds to the A5 components of the SO(5)/SO(4) 5D gauge
fields. There are various ways to obtain the fermion content of the SM. In [17], two possibilities
are presented, labeled MCHM5 (where MCHM stands for Minimal Composite Higgs Model)
and MCHM10 , where the SM fermions are contained in the 5 and 10 representations of SO(5),
respectively.
Let us first examine in detail the MCHM5 model. Each SM generation of quarks is identified
with zero modes of the following bulk multiplets (where BCs are indicated explicitly) [17],

q (+)
q (+)
q
q
(2, 2)L1 = q 1L(++)
(2, 2)R1 = q 1R()
1L
1R
q1 =
,
q1
q1
(1, 1)L ()
(1, 1)R (++)

(2, 2)uL (+) (2, 2)uR (+)
,
u =
(1, 1)uL (+) (1, 1)uR (+)
380

q2 =
q
2L (++)
(+)
q2L

d =
(2, 2)L2 =
(2, 2)R2 =
(1, 1)dL (+)
2R ()
(+)
q2R

(1, 1)L2 ()
(2, 2)dL (+)
q
(1, 1)R2 (++)
(2, 2)dR (+)

,
(1, 1)dR (+)
(63)
where q1 , u (q2 , d ) transform as 52/3 (51/3 ) multiplets of SO(5) U (1)X . The fields are
grouped within the multiplets in representations of SO(4) SU(2)L SU(2)R , using the decomposition 5 = 4 1 = (2, 2) (1, 1). For each generation there is an additional right-handed
q
(chirality under 4D Lorentz group) doublet of SU(2)L , qR = (2, 0)R , localized on the UV brane.
These boundary quarks transform as 3s of SU(3) and have Y = 1/6. It is suggested in [17] that
the leptons be embedded in a similar manner to the quarks but with differing U (1)X charges.
Doing this we have the following bulk multiplets,

l1 =
(2, 2)lL1 =
(+)
l1L
l1L (++)
(2, 2)lR1 =
(+)
l1R
l1R ()

,
(1, 1)lR1 (++)

(2, 2)L (+) (2, 2)R (+)
,
=
(1, 1)L (+) (1, 1)R (+)

l (++)
l ()
(2, 2)lL2 = l 2L(+)
(2, 2)lR2 = l 2R(+)
2L
2R
,
l2 =
(1, 1)lL2 ()
(1, 1)lR2 (++)

(2, 2)eL (+) (2, 2)eR (+)
,
e =
(1, 1)eL (+) (1, 1)eR (+)
(1, 1)lL1 ()
(64)
where l1 , (l2 , e ) transform as 50 (51 ) multiplets of SO(5) U (1)X . In analogy with the
quarks, for each generation of leptons there is an additional right-handed doublet of SU(2)L ,
lR = (2, 0)lR , localized on the UV brane. These boundary leptons have Y = 1/2. We find
that for the fermion content described above there are no anomalies for either H0 or H1 in the
MHCM5 . Consequently, this particular version of the MCHM generates the correct spectra of
SM gauge and Higgs bosons, without recourse to ChernSimons terms. We have also computed
the anomalies for the MCHM10 and various other warped models of SM EWSB, both with and
without a Higgs. In many of these models the full content of the fermionic sector is somewhat
ambiguous. However, in the same way as we have done for the MCHM5 we can write down a
plausible fermionic sector and calculate the resulting fermionic anomalies. We find that in the
majority of these models there are non-vanishing brane-localized fermion anomalies which need
to be cancelled in order to get the correct description of the SM at low energies. Two possible ways to cancel these anomalies are either to add ChernSimons terms to the d = 5 theory,
and/or, more generally, to change the fermion content of the model; such changes will affect the
low-energy phenomenology, which we hope to explore in future work [19].
Acknowledgements
We thank J. March-Russell for useful discussions. This work was partially supported by the
EU FP6 Marie Curie Research and Training Network UniverseNet (HPRN-CT-2006-035863).
381
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.

E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
N. Arkani-Hamed, M. Porrati, L. Randall, JHEP 0108 (2001) 017, hep-th/0012148.
R. Rattazzi, A. Zaffaroni, JHEP 0104 (2001) 021, hep-th/0012248.
M. Perez-Victoria, JHEP 0105 (2001) 064, hep-th/0105048.
L.D. Faddeev, S.L. Shatashvili, Theor. Math. Phys 60 (1985) 770;
C.G. Callan, J.A. Harvey, Nucl. Phys. B 250 (1985) 427.
N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 516 (2001) 395, hep-th/0103135.
C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073.
R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, Phys. Rev. D 66 (2002) 024025, hep-th/0203039.
G. von Gersdorff, M. Quiros, Phys. Rev. D 68 (2003) 105002, hep-th/0305024.
C.A. Scrucca, M. Serone, Int. J. Mod. Phys. A 19 (2004) 2579, hep-th/0403163.
A. Boyarsky, O. Ruchayskiy, M. Shaposhnikov, Phys. Rev. D 72 (2005) 085011, hep-th/0507098;
A. Boyarsky, O. Ruchayskiy, M. Shaposhnikov, Phys. Lett. B 626 (2005) 184, hep-ph/0507195.
T. Hirayama, K. Yoshioka, JHEP 0401 (2004) 032, hep-th/0311233.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
G. t Hooft, in: G. t Hooft, et al. (Eds.), Nato Advanced Study Institutes Series: Series B, Physics, vol. 59, Plenum,
New York, USA, 1980, 438 p.
See e.g. R. Contino, Y. Nomura, A. Pomarol, Nucl. Phys. B 671 (2003) 148, hep-ph/0306259;
K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165, hep-ph/0412089;
K. Agashe, R. Contino, L. Da Rold, A. Pomarol, Phys. Lett. B 641 (2006) 62, hep-ph/0605341.
R. Contino, L. Da Rold, A. Pomarol, hep-ph/0612048.
See e.g. C. Csaki, C. Grojean, L. Pilo, J. Terning, Phys. Rev. Lett. 92 (2004) 101802, hep-ph/0308038;
Y. Nomura, JHEP 0311 (2003) 050, hep-ph/0309189;
G. Cacciapaglia, C. Csaki, G. Marandella, J. Terning, Phys. Rev. D 75 (2007) 015003, hep-ph/0607146;
G. Cacciapaglia, C. Csaki, G. Marandella, J. Terning, JHEP 0702 (2007) 036, hep-ph/0611358.
B. Gripaios, J. March-Russell, S.M. West, in preparation.
G. Panico, A. Wulzer, hep-th/0703287.
See e.g. C. Csaki, J. Hubisz, P. Meade, hep-ph/0510275.
B. Gripaios, Nucl. Phys. B 768 (2007) 157, hep-ph/0611218.
C. Csaki, C. Grojean, J. Hubisz, Y. Shirman, J. Terning, Phys. Rev. D 70 (2004) 015012, hep-ph/0310355.
C.T. Hill, Phys. Rev. D 73 (2006) 085001, hep-th/0601154;
C.T. Hill, Phys. Rev. D 73 (2006) 126009, hep-th/0603060.
M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
S. Groot Nibbelink, H.P. Nilles, M. Olechowski, M.G.A. Walter, Nucl. Phys. B 665 (2003) 236, hep-th/0303101.
L.E. Ibanez, R. Rabadan, A.M. Uranga, Nucl. Phys. B 542 (1999) 112, hep-th/9808139;
C.A. Scrucca, M. Serone, JHEP 9912 (1999) 024, hep-th/9912108;
C.A. Scrucca, M. Serone, M. Trapletti, Nucl. Phys. B 635 (2002) 33, hep-th/0203190.
J. Preskill, Ann. Phys. 210 (1991) 323.
J. Wess, B. Zumino, Phys. Lett. B 37 (1971) 95;
E. Witten, Nucl. Phys. B 223 (1983) 422.
D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96, hep-th/9804058;
G. Chalmers, H. Nastase, K. Schalm, R. Siebelink, Nucl. Phys. B 540 (1999) 247, hep-th/9805105.
H.L. Verlinde, Nucl. Phys. B 580 (2000) 264, hep-th/9906182;
J. de Boer, E.P. Verlinde, H.L. Verlinde, JHEP 0008 (2000) 003, hep-th/9912012;
C. Csaki, J. Erlich, T.J. Hollowood, J. Terning, Phys. Rev. D 63 (2001) 065019, hep-th/0003076;
A. Lewandowski, M. Redi, Phys. Rev. D 68 (2003) 044012, hep-th/0305013.
A. Lewandowski, M.J. May, R. Sundrum, Phys. Rev. D 67 (2003) 024036, hep-th/0209050.
Geometrically induced metastability and holography

Mina Aganagic a, , Christopher Beem a , Jihye Seo b , Cumrun Vafa b,c
a University of California, Berkeley, CA 94720, USA
b Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, USA
c Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Received 16 August 2007; accepted 21 August 2007

Available online 6 September 2007
Abstract
We construct metastable configurations of branes and anti-branes wrapping 2-spheres inside local Calabi
Yau manifolds and study their large N duals. These duals are CalabiYau manifolds in which the wrapped
2-spheres have been replaced by 3-spheres with flux through them, and supersymmetry is spontaneously
broken. The geometry of the non-supersymmetric vacuum is exactly calculable to all orders of the t Hooft
parameter, and to the leading order in 1/N . The computation utilizes the same matrix model techniques that
were used in the supersymmetric context. This provides a novel mechanism for breaking supersymmetry in
the context of flux compactifications.
1. Introduction
One of the central questions currently facing string theory is how to break supersymmetry in
a controllable way. The most obvious ways to break it typically lead to instabilities signaled by
the appearance of tachyons in the theory. One would like to find vacua in which supersymmetry
is broken, but stability is not lost. It seems difficult (or impossible, at present) to obtain exactly
stable non-supersymmetric vacua from string theory. Therefore, the only candidates would appear to be meta-stable non-supersymmetric vacua. This idea has already been realized in certain
models (see [1] for a review and the relevant literature). More recently, the fact that metastable
vacua are also generic in ordinary supersymmetric gauge theories [2] has added further motivation for taking this method of breaking supersymmetry seriously within string theory. Potential
E-mail address: mina@math.berkeley.edu (M. Aganagic).

doi:10.1016/j.nuclphysb.2007.08.018
M. Aganagic et al. / Nuclear Physics B 789 (2008) 382412
383
realizations of such metastable gauge theories have been considered in string theory [3] (see
also [4]).
The aim of this paper is to study an alternative approach to breaking supersymmetry via
metastable configurations, as suggested in [5]. In this scenario, we wrap branes and anti-branes
on cycles of local CalabiYau manifolds, and metastability is a consequence of the CalabiYau
geometry. In a sense, this is a geometrically induced metastability. The branes and the anti-branes
are wrapped over 2-cycles which are rigid and separated. In order for the branes to annihilate,
they have to move, which costs energy as the relevant minimal 2-spheres are rigid. This leads
to a potential barrier due to the stretching of the brane and results in a configuration which is
metastable. It is particularly interesting to study the same system at large N , where we have a
large number of branes and anti-branes. In this case, it is better to use a dual description obtained
via a geometric transition in which the 2-spheres are shrunk, and get replaced by 3-spheres
with fluxes through them. The dual theory has N = 2 supersymmetry, which the flux breaks
spontaneously. If we have only branes in the original description, then the supersymmetry is
broken to an N = 1 subgroup. With only anti-branes present, we expect it to be broken to a
different N = 1 subgroup, and with both branes and anti-branes, the supersymmetry should
be completely broken. The vacuum structure can be analyzed from a potential which can be
computed exactly using topological string theory [6] or matrix models [7].
Unlike the cases studied beforeinvolving only braneswith branes and anti-branes present,
we expect to find a meta-stable vacuum which breaks supersymmetry. We will find that this is
the case, and moreover this leads to a controllable way of breaking supersymmetry at large N
where to all orders in the t Hooft coupling, but to leading order in the 1/N expansion, we can
compute the geometry of the vacua and the low energy Lagrangian.
The organization of this paper is as follows: In Section 2 we review the case where we have
a single stack of branes and extend it to the case where we have a single stack of anti-branes. In
Section 3 we discuss the case with more than one S 2 in the geometry, and then specialize to the
case in which there are only two. We will explain how if we have only branes or only anti-branes,
supersymmetry is not broken, whereas if we have branes wrapped on one S 2 and anti-branes on
the other, supersymmetry is spontaneously broken. In Section 4 we estimate the decay rate of
the metastable vacuum. In Section 5 we conclude with open questions and suggestions for future
work.
2. Branes and anti-branes on the conifold
Let us begin by recalling the physics of N D5 branes on the conifold singularity. We consider
type IIB string theory with the branes wrapping the S 2 obtained by resolving the conifold singularity. The local geometry of the CalabiYau threefold is a P1 with normal bundle given by the
sum of two line bundles
O(1) + O(1) P1 .
At low energies, the field theory on the space-filling D5 branes is a pure U (N ) gauge theory
with N = 1 supersymmetry. That the theory has N = 1 supersymmetry follows from the fact
that string theory on the background of local CalabiYau manifolds preserves N = 2, and the Dbranes break half of it. In particular, there are no massless adjoint fields, which follows from the
important geometric fact that the P1 which the D5 brane wraps is isolated (its normal bundle is
O(1) + O(1), which does not have a holomorphic section). In other words, any deformation
384
Fig. 1. The geometry near a resolved conifold. Moving away in the normal direction, the volume of the wrapped 2-cycle
(represented here as an S 1 ) must increase.
of the P1 in the normal direction increases its volume, and so corresponds to a massive adjoint
field, at best (see Fig. 1).
The N = 1 pure YangMills theory on the brane is expected to be strongly coupled in the
IR, leading to gaugino bilinear condensation, confinement, and a mass gap. There are N massive
vacua corresponding to the gaugino superfield getting an expectation value:

2i
2ik
3
S = 0 exp
(2.1)
exp
, k = 1, . . . , N.
N
N
In the future we will suppress the phase factor which distinguishes the N vacua. Above, is the
bare gauge coupling constant defined at the cutoff scale 0
(0 ) =
and S =
1
32 2
4
,
i
2
gYM 2 (0 )
Tr W W . Recalling that the gauge coupling in this theory runs as
3N
,
2i(0 ) = log
0
(2.2)
where is the strong coupling scale of the theory, we can also write (2.1) as
S = 3 .
(2.3)
The theory has an anomalous axial U (1)R symmetry which rotates the gauginos according to
ei , so
S Se2i .
The anomaly means that this is a symmetry of the theory only provided the theta angle shifts
+ 2N.
(2.4)
Since the theta angle shifts, the tree level superpotential Wtree = S is not invariant under the
R-symmetry, and in the quantum theory additional terms must be generated to correct for this.
Adding the correction terms produces the effective VenezianoYankielowicz superpotential [8],

1
NS log S/30 1
2i
whose critical points are (2.1).
W(S) = S +
(2.5)
385
One can also understand the generation of this superpotential from the viewpoint of the large
N holographically dual theory [5]. This configuration of branes wrapping a P1 is dual to a closed
string theory on the CalabiYau manifold obtained by a geometric transition which replaces the
wrapped P1 with a finite sized S3 . In the dual theory, the 5-branes have disappeared, and in their
place there are N units of RamondRamond flux

H =N
A
through the 3-cycle A corresponding to the S 3 . Here

H = H RR + H NS ,
where is the type IIB dilaton-axion, = C0 +
dual B cycle

H = .
i
gs .
There are also fluxes turned on through the
These fluxes generate a superpotential [9,10]:

W = H ,
where is the holomorphic (3, 0) form on the CalabiYau manifold. This can be written in
terms of the periods

S = ,
F0 =
S
A
as
F0 .
(2.6)
S
Above, F0 is the prepotential of the N = 2 U (1) gauge theory which is the low-energy effective
theory of type IIB string theory on this geometry before turning on fluxes.
For our case, the B-period was computed in [5]
W(S) = S + N

1
F0 =
S log S/30 1
S
2i
and so (2.6) exactly reproduces the effective superpotential (2.5) which we derived from gauge
theory arguments. Moreover, the superpotential is an F-term which should not depend on the
cutoff 0 , so the flux has to run (this agrees with the interpretation of 0 as an IR cutoff in the
conifold geometry which regulates the non-compact B-cycle)
3N
2i(0 ) = log
.
0
In the dual gauge theory, this corresponds to the running of the gauge coupling in the low energy theory. This theory and the duality were studied from a different perspective in [11] (see
also [12]).
386
2.1. The anti-brane holography

Now consider replacing the D5 branes with anti-D5 branes wrapping the P1 . It is natural
that the physics in the presence of the two types of branes should be identical. In particular,
in the open string theory, we again expect gaugino condensation, a mass gap and confinement.
We conjecture that N anti-branes wrapping the P1 are also holographically dual to the conifold
deformation with flux through the S 3 . In fact, we have no choice but to require this as it is the
result of acting by CPT on both sides of the duality. On the open string side, we replace the
branes with anti-branes, and on the closed string dual we have N units of flux through the S 3 ,
but N is now negative. In other words, the superpotential is still given by (2.5), but with N < 0.
At first sight, this implies that there is a critical point as given by (2.1), but now with N
negative. However, this cannot be right, since S, which is the size of the S 3 , would grow without
8 2
bound as we go to weak coupling S exp( |N|g
2 ). This is clearly unphysical, and moreover
YM
the description of the conifold breaks down when S is larger than the cutoff 30 in the dual closed
string geometry.
To see what is going on, recall that on the open string side the background has N = 2 supersymmetry, and adding D5 branes preserves an N = 1 subset of this, while adding anti-branes
preserves an orthogonal N = 1 subset. By holography, we should have the same situation on the
other side of the duality. Namely, before turning on flux, the background has N = 2 supersymmetry, and turning on flux should break this to N = 1. But now holography implies that depending
on whether we have branes or anti-branes in the dualso depending on the sign of N we
should have one or the other N = 1 subgroup of the supersymmetry preserved. It is clear that
the superpotential (2.5) has been adapted to a superspace in which the manifest N = 1 supersymmetry is the one preserved by branes, and hence is not well adapted for the supersymmetry
of the anti-branes. Nevertheless, the theory with negative flux should somehow find a way to
capture the supersymmetry, as string holography predicts! We will now show that this is indeed
the case.
The vacua of the theory are clearly classified by the critical points of the physical potential V
of the theory:
S V = 0,
where
V = g S S |S W|2 .
Above, gS S is the Khler metric for S. In this case, the theory has softly broken N = 2 supersymmetry, and so g is given in terms of the prepotential as in the N = 2 case
gS S = Im( ),
where

1
log S/30 .
2i
With superpotential W as given in (2.6), the effective potential becomes
(S) = S2 F0 =
V=
2i
| + N |2 .
( )
387
It is easy to see that the critical points are at

S V =
2i
3 F0 ( + N )( + N ) = 0.
( )2 S
This has two solutions. The first is at

+ N = 0
(2.7)
which solves W = 0, and corresponds to (2.1). It is physical when N is positive. The second
critical point is at
+ N = 0.
(2.8)
Note that in this vacuum, W = 0. In terms of S, it corresponds to

2i
(
0)
S = 0 3 exp
|N|
(2.9)
and is the physical vacuum when N is negative (i.e., where we have |N | anti-branes).
So how can it be that even though W = 0, supersymmetry is nevertheless preserved? Toward
understanding this, recall that before turning on flux, the closed string theory has N = 2 supersymmetry with one N = 2 U (1) vector multiplet A. Adding D5 branes in the original theory
corresponds to turning on positive flux, which forces this to decompose into two N = 1 supermultiplets: a chiral multiplet S containing S and its superpartner , and a vector multiplet W
containing the U (1) gauge field (coming from the 4-form potential decomposed in terms of the
harmonic 3-form on S 3 and a 1-form in 4d) and the gaugino :

A = S, W ,
where
S = S + + ,

i
W = + F + .
2
The Lagrangian in N = 1 superspace is given by
L = L0 + LW ,
where
L0 = Im

d 2 d 2 Si
F0
+
Si

d 2
1 2 F0
W Wj
2 Si Sj i
is the action for an N = 2 supersymmetric theory with prepotential F0 , and the superpotential
term is

LW = d 2 W(S) + c.c.,
where the superpotential W(S) is given by (2.6) and is repeated here for the readers convenience
W(S) = S + N
F0 .
S
388
The puzzle is now what happens in the anti-D5 brane case. Since the Lagrangian with flux apparently has only N = 1 supersymmetry preserved by the D5 branes for any value of N , positive
or negative, how is it possible that the anti-brane preserves a different N = 1? One might guess
that despite the flux, the Lagrangian L actually preserves the full N = 2 supersymmetry. This
is too much to hope for. In particular, turning on flux should be holographically dual to adding
in the branes, which does break half of the N = 2 supersymmetry of the background. Instead,
it turns out that the flux breaks N = 2 supersymmetry in a rather exotic way. Namely, which
N = 1 is preserved off-shell turns out to be a choice of a gauge: we can write the theory in a
way which makes either the brane or the anti-brane N = 1 supersymmetry manifest, no matter
what N is. On shell however we have no such freedom, and only one N = 1 supersymmetry can
be preserved. Which one this is depends only on whether the flux is positive or negative, and not
on the choice of the N = 1 supersymmetry made manifest by the Lagrangian.
To see how all this comes about, let us try to make the spontaneously broken N = 2 supersymmetry manifest. The N = 2 vector multiplet A is really a chiral multiplet, satisfying the
N = 2 chiral constraint D i A = 0,
A = S + i i + i j Xij +

1
ij i j F + .
2
(2.10)
Here i, j = 1, 2 are the SU(2)R indices, and i is a doublet of fermions:

.
=
The auxiliary fields Xij of each N = 2 chiral superfield form an SU(2)R triplet, satisfying a
reality constraint
X ij = il j k X lk Xij .
(2.11)
In particular, X 11 = X 22 = X22 , and so on.1

The action L can be written in terms of N = 2 superfields, where turning on fluxes in the
geometry corresponds to giving a vacuum expectation value to some of the N = 2 F-terms [5,
13,14]. Our presentation here follows closely [14]. Namely, consider the Lagrangian

2
2
Im
d 1 d 2 F0 (A) + Xij E ij + X ij Eij ,
(2.12)
where E ij is the triplet of FayetIliopoulos terms, with same properties as X has. Since the Xij
transform by total derivatives, the FI term Xij E ij preserves the N = 2 supersymmetry. This will
match L precisely if we set
E 11 = = E 22 ,
E 12 = 0
and moreover, we give Xij a non-zero imaginary part

Xij Xij + iNij ,
(2.13)
1 More precisely, A is a reduced N = 2 chiral multiplet, meaning it satisfies an additional constraint: D ij D A

ij
where Dij = Di D , and 2 is the standard Laplacian. The reducing constraint says that 2 il j k X lk = 2 X ij .
2 A,
j
This implies that we can shift X by a constant imaginary part that does not satisfy (2.11).
389
where
N11 = 0,
N22 = 2N,
N12 = 0.
(2.14)
It is easy to see from (2.10) that to decompose A in terms of N = 1 multiplets, we can simply
expand it in powers of 2
A(1 , 2 ) = S(1 ) + 2 W (1 ) + 2 2 G(1 ),
(2.15)
where the chiral multiplet G is given by (see, for example [15])2

i1 1 , 1 ) = X22 + .
G(y, 1 ) = d 2 1 S(y
Then, plugging in the vacuum expectation values above in (2.12) and integrating over 2 , we
recover the N = 1 form of L. More precisely (this will get a nice interpretation later) we recover
2 . Note that by shifting X we have turned on
it, with the addition of a constant term 8N/gYM
2
a non-zero F-term in the direction off-shell. This breaks N = 2 supersymmetry, leaving the
Lagrangian with only N = 1 supersymmetry along 1 direction.
Consider now the vacua of the theory.3 Let us denote the full F-term as
X ij = Xij + iNij .
Then it is easy to see that X gets an expectation value,
X 11 =
2i
( + N )

(2.16)
and
2i
( + N ).
(2.17)

Now, recall that the physical vacua depend on whether the flux is positive or negative. In particular, in the brane vacuum, where N is positive, we have + N = 0, so
X 22 =
X 11 = 0,
X 22 = 0.
Now, since the supersymmetry variations of the fermions are

i = i X ij j + .
It follows immediately that in the brane vacuum
= 0,
= 0
2 Explicitly,
2 2 G(1 ) = X22 2 2 +

2
1
1
ij i j 2 +
ij i j 2 S.
2
3!
See for example [16].

3 Explicitly, in this language, the F-term potential becomes
1 ij
1
Xij X X ij X ij + X ij E ij + X ij Eij ,
2i
2i
where indices are always raised and lowered with tensor, e.g., X ij = ik j l X kl . Moreover, using reality properties
it is easy to see that X ij = X ij i(Nij + N ij ), and the result follows.
of X,
390
and is the Goldstino. The unbroken supersymmetry pairs up S and into a chiral field S and
and the gauge field into W , as in (2.15).
By contrast, in the anti-brane vacuum, + N = 0, so
X 11 = 0,
X 22 = 0.
Correspondingly, now is the Goldstino:

= 0,
= 0.
and
Now the unbroken supersymmetry corresponds to pairing up S and into a chiral field S,
is the partner of the gauge field in W . In other words, now, it is natural to write A as
W )
A = (S,
or, more explicitly:
2 ) + W (2 ) + 1 1 G(
2 ).
A(1 , 2 ) = S(
1
(2.18)
It should now be clear how it comes about that even though the Lagrangian L has only N = 1
supersymmetry, depending on whether the flux N is positive or negative, we can have different
N = 1 subgroups preserved on shell. The supersymmetry of the Lagrangian was broken since
the flux shifted X as in (2.13). More precisely, supersymmetry was broken only because this
shift of X could not be absorbed in the field redefinition of X, consistent with (2.11). Off-shell,
X is allowed to fluctuate, but because the shift is by iNij , which is not of the form (2.11), its
fluctuations could not be absorbed completely and supersymmetry really is broken. However the
shift of X by Nij in (2.14) is indistinguishable from shifting X by
N11 = 2N,
N22 = 0,
N12 = 0,
(2.19)
since the difference between the two shifts can be absorbed into a redefinition of the fields.
This would preserve a different N = 1 subgroup of the N = 2 supersymmetry, the one that is
natural in the anti-brane theory. Correspondingly, in all cases, N = 2 supersymmetry is broken
to N = 1 already at the level of the Lagrangian. However, which N = 1 is realized off-shell is a
gauge choice.
Next we compute the masses of bosonic and fermionic excitations around the vacua. The
relevant terms in the N = 2 Lagrangian are:

1
d 4 S3 F0 i i j j X kl k l .
2
In the present context, we can simplify this as
1 3
1
F0 (1 1 )X 22 + S3 F0 (2 2 )X 11 .
2 S
2
This gives fermion masses4
2i
( + N )S3 F0 ,
( )2
2i
( + N )S3 F0 .
m =
( )2
m =
4 This is the physical mass, with canonically normalized kinetic terms.
391
For comparison, the mass of the glueball field S is5

1/2 3
2
F0 .
| + N |2 + | + N |2
mS =
S
2
| |
As a check, note that in the brane vacuum (2.7), is indeed massless as befits the partner of the
gauge field (since the original theory had N = 2 supersymmetry, here is in fact a Goldstino!).
Moreover, it is easy to see that the masses of and S agree, as they should,
1
N2
m = 0.
= |m |,
2 |S|| Im()|
Now consider the case of anti-branes where, in the holographic dual, we have negative flux.
This corresponds to the vacuum (2.8). The gauge field A is still massless and S is massive. From
(2.8), it follows that in this vacuum it is which is massless, and becomes massive. Moreover,
the mass of is the same as for S
|mS | =
1
N2
m = 0.
= |m |,
2 |S|| Im()|
This is all beautifully consistent with holography!
To summarize, the anti-brane vacuum preserves different supersymmetry than the brane vacuum, and the same is true in the large N dual. In other words the anti-brane/negative flux has
oriented the N = 1 supersymmetry differently, as is expected based on the holographic duality.
Let us now try to see if we can understand the anti-brane gluino condensate (2.9) directly
from the gauge theory. Before adding in branes, the background has N = 2 supersymmetry. The
corresponding supercharges are two Weyl fermions Q, Q that transform as a doublet under the
SU(2)R symmetry of the theory. In either vacuum with the branes, only the U (1)R subgroup
of the SU(2)R symmetry is preserved. If adding in D-branes preserves Q, then anti-D branes
Since Q and Q transform oppositely under the U (1)R symmetry, in going from
will preserve Q.
brane to anti-brane, the chirality of the world-volume fermions flips.
that is an honest F-term with respect to the supersymmetry preserved
The superpotential W
by the anti-brane theory, and which reproduces (2.9), is
|mS | =

1
= S
W
+
N S log S/30 1 ,
2i
where S is the anti-brane gaugino condensate, and N is negative. This reflects the fact that, if
we keep the background fixed, then transforms under the R symmetry still as (2.4), but the
gauginos transform oppositely from before. The rest is fixed by holographyin other words,
that is the F-term in this
is the good chiral field which has as a component. Since it is W
case, it is this, and not W, that should be independent of scale for the anti-brane, and hence
should run as

3N
2i(0 ) = log
.
0
Note that this is consistent with turning on the other F-term vev (2.19), which directly preserves
the anti-brane supersymmetry.
5 It is important to note here that the kinetic terms for S are fixed for us by the string large N duality, and N = 2
supersymmetry. They differ from the canonical kinetic terms of [8]. We are writing here the physical masses, in the
basis where all the fields have canonical kinetic terms.
392
Finally, consider the value of the potential in the minimum. At the brane vacuum, supersymmetry is unbroken, and
V brane = 0.
At the anti-brane vacuum, with negative N , we have:
V anti-brane = 16
N
|N |
= 16 2 .
2
gYM
gYM
This treats branes and anti-branes asymmetrically. It is natural then to shift the zero of the potential and have
N
V = V + 8 2
gYM
which leads to the minimum value
|N |
Vanti-brane = 8 2 = Vbrane .
gYM
This is equivalent to the inclusion of the constant term which came from the magnetic FI term in
the N = 2 theory. Note that this corresponds to the tension of |N | (anti-)branes which is predicted
by holography! We thus find a beautiful confirmation for the stringy anti-brane holography.
3. Geometric metastability with branes and anti-brane
There are many ways to naively break supersymmetry in string theory, but they typically lead
to instabilities. The simplest example of such a situation is found in type IIB string compactifications on a CalabiYau, where one introduces an equal number of D3 branes and anti-branes
filling the spacetime, but located at points in the CalabiYau. Supersymmetry is clearly broken,
since branes and anti-branes preserve orthogonal subsets of supersymmetry. This does not typically lead to metastability, however, as one can move branes to where the anti-branes are located,
and such a motion would be induced by an attractive force. This is then similar to a familiar
problem in classical electrostatics, and is plagued by the same difficulty as one has in finding
locally stable equilibrium configurations of static electric charges.
One way to avoid such an obstacle, as was suggested in [5], is instead of considering branes
and anti-branes occupying points in the CalabiYau, to consider wrapping them over minimal
cycles inside the CalabiYau. Moreover, we choose the minimal cycles to be isolated. In other
words we assume there is no continuous deformations of the minimal cycles. We wrap branes
over a subset C1 of such cycles and we wrap the anti-branes over a distinct subset C2 . In the
case of compact CalabiYau, it is necessary that the class [C1 ] + [C2 ] = 0, so that there is no
net charge (this condition can be modified if we have orientifold planes). Because the branes and
anti-branes cannot move without an increase in energy, these will be a metastable equilibrium
configurations (see Fig. 2). This is true as long as the cycles C1 and C2 are separated by more
than a string scale distance, so that the geometric reasoning remains valid (when they are closer
there are tachyon modes which cause an instability).
Here we will consider non-compact examples of this scenario. This decouples the lower dimensional gravity from the discussion, and moreover, the condition that the net class [C1 ] + [C2 ]
be zero is unnecessary as the flux can go to infinity. In particular, we will consider type IIB
strings on a non-compact CalabiYau threefold where we can wrap 4d spacetime filling D5 and
393
Fig. 2. Before anti-branes and branes can annihilate one other, they will have to move and meet somewhere, and thus
they will have to increase their volume due to the CalabiYau geometry.
anti-D5 branes over different isolated S 2 s in the CalabiYau. The local geometry of each S 2
will be the resolved conifold reviewed in the previous section. The only additional point here is
that we have more than one such S 2 in the same homology class. Moreover, we will consider
the large N limit of such brane/anti-brane systems and find that the holographically dual closed
string geometry is the same one as in the supersymmetric case with just branes, except that some
of the fluxes will be negative. This leads, on the dual closed string side, to a metastable vacuum
with spontaneously broken supersymmetry.
We will first describe general geometries of this kind which support meta-stable configurations
with both D5 and anti-D5 branes. The relevant CalabiYau geometries turn out to be the same
ones studied in [6], which led to a non-perturbative formulation of the dual geometry in terms of
matrix models [7]. The new twist is that we now allow not just branes, but both branes and antibranes to be present. We will describe the holographically dual flux vacua. We then specialize
to the case of just two S 2 s with branes wrapped over one S 2 and anti-branes wrapped over the
other, and study it in more detail.
3.1. Local multi-critical geometries
Consider a CalabiYau manifold given by
uv = y 2 + W (x)2 ,
(3.1)
where
W (x) = g
(x ak ).
k=1
If, for a moment, we set g = 0, then this is an A1 ALE singularity at every point in the x plane.
One can resolve this by blowing up, and this gives a family of P1 , i.e., holomorphic 2-spheres,
parameterized by x. Let us denote the size of the blown up P1 by r. In string theory, this is
complexified by the B NS -field.6 Turning g back on lifts most of the singularities, leaving just n
isolated P1 s at x = ak . Of course, the S 2 still exists over each point x, but it is not represented
6 On the world volume of the D5 brane B
NS gets mixed up with the RR potential, so more precisely by r/gs we will
mean the complex combination of the Khler modulus with B RR + B NS . To not have the dilaton turned on, we will in
fact keep only the later, setting the geometric size to zero [11].
394
by a holomorphic P1 . In other words, its area is not minimal in its homology class. We have

1/2
A(x) = |r|2 + |W |2
(3.2)
,
where A(x) denotes the area of the minimal S 2 as a function of x. This can be seen by the fact
that the equation for each x is an ALE space and the above denotes the area of the smallest S 2
in the ALE geometry for a fixed x. Even though the minimal P1 s are now isolated, i.e., at points
where W = 0, they are all in the same homology class, the one inherited from the ALE space.7
Note also that the first term, |r|2 , does not depend on x.
One can now consider wrapping some number of branes Nk , k = 1, . . . , n, on each P1 . The
case when all Nk s are positive and all the branes are D5 branes was studied in [6]. In that case,
the gauge theory on the branes is an N = 1 supersymmetric U (N ) gauge theory with an adjoint
matter field and superpotential given by
Tr W (),
(3.3)
W (x)
where W is the same polynomial whose derivative is given by

in the defining equation
of CalabiYau [6,17]. The eigenvalues of are identified with positions of the D5 brane on the
x-plane. Here
N=
Nk
k
and the choice of distribution of D5 branes among the critical points ak corresponds to the choice
of a Higgs branch of this supersymmetric gauge theory where, the gauge group is broken
U (N )
U (Nk ).
k=1
In the low energy limit we have N = 1 YangMills theories U (Nk ) with field corresponding
to a massive adjoint matter for each. Note that (3.3) is consistent with (3.2) in the large r limit,
because wrapping a brane over P1 and considering its energy as a function of x, it is minimized
along with the area at points where W (x) = 0. Furthermore, it is clear from this that the effective
coupling constant gYM of the four-dimensional gauge theory living on the brane is the area of
the minimal P1 s times 1/gs :
1
2
gYM
|r|
.
gs
Below, we will study what happens when some of the P1 s are wrapped with D5 branes
and others with anti-D5 branes. By making the vacua ai very widely separated, the branes and
the anti-branes should interact very weakly. Therefore, we still expect to have an approximate
N = 1 supersymmetric gauge theory for each brane, with gaugino condensation and confinement
at low energies, just as discussed in the previous section. However, because supersymmetry is
broken and there are lower energy vacua available where some of the branes annihilate, the
system should be only meta-stable. Note that the fact that the P1 s are isolated but in the same
class is what guarantees metastability. For the branes to annihilate the anti-branes, they have to
7 As explained in [9], the parameters g and a which enter (3.1) and W are not dynamical fields, but enter in specifying
i
the theory. This is possible because the CalabiYau is non-compact.
395
1 |dW |2 A(x) is depicted along a real

Fig. 3. The brane theory with a tree level physical potential gs V (x) = |r| + 2r
line in the x-plane for the case where W has two critical points. The potential reflects the CalabiYau geometry, and the
leading term represents the brane tension.
climb the potential depicted in Fig. 3. This should be accurate description of the potential when
the branes are far away from each other, where the effect of supersymmetry breaking is small.
When the branes and anti-branes are very close togetherfor example when they are within a
string distancethere would be tachyon in the theory and the above potential will not be an
accurate description. Nevertheless it is clear that the minimum of a brane and anti-brane system
is realized when they annihilate each other. We have thus geometrically engineered a metastable
brane configuration which breaks supersymmetry. We will discuss aspects of the open string
gauge dynamics of this configuration in Section 3.6. As we will discuss in that section, unlike
the supersymmetric case, there seems to be no simple field theory with a finite number of degrees
of freedom which captures the brane/anti-brane geometry. Of course, one can always discuss it
in the context of open string field theory.
We have a control parameter for this metastability, which also controls the amount of supersymmetry breaking, which is related to the separation of the critical points ai . The farther
apart they are, the more stable our system is. We will discuss stability and decay rate issues in
Section 4. In the next subsection, we study the large N holographic dual for this system.
3.2. The large N dual description
The supersymmetric configuration of branes for this geometry was studied in [6], where a
large N holographic dual was proposed. The relevant CalabiYau geometry was obtained by
a geometric transition of (3.1) whereby the P1 s are blown down and the n resulting conifold
singularities at x = ak are resolved into S 3 s by deformations of the complex structure. It is
given by
uv = y 2 + W (x)2 + fn1 (x),
(3.4)
where fn1 (x) is a degree n 1 polynomial in x. As explained in [6], the geometry is effectively described by a Riemann surface which is a double cover of the x plane, where the two
sheets come together along n cuts near x = ak (where the P1 s used to be). The A and B cycles of the CalabiYau project to 1-cycles on the Riemann surface, as in Fig. 4. The geometry is
characterized by the periods of the (3, 0) form ,

Sk F0 = .
Sk = ,
Ak
Bk
396
Fig. 4. The multi-critical CalabiYau geometry projected to the x-plane. The cut structure reflects the branching of y
over the x plane for u = v = 0.
The complex scalars Sk are the scalar partners of the n U (1) vector multiplets under the N = 2
supersymmetry, and F0 (S) is the corresponding prepotential. As before, while (3.4) depends also
on ai and g, the latter are just fixed parameters.
If, before the transition, all of the P1 s were wrapped with a large number of branes, the
holographically dual type IIB string theory is given by the (3.4) geometry, where the branes from
before the transition are replaced by fluxes

H = Nk ,
H = .
Bk
Ak
We want to conjecture that this duality holds whether or not all the Nk 0. In fact, the discussion
of the previous section shows that if all the Nk 0 the duality should continue to hold by CPT
conjugation. Our conjecture is that it also holds as when some Nk are positive and some negative.
The flux numbers Nk will be positive or negative depending on whether we had D5 branes or
anti-D5 branes wrapping the kth P1 before the transition. The flux through the Bk cycles will
correspond to the bare gauge coupling constant on the D-branes wrapping the corresponding P1 ,
and is the same for all k as the P1 s are all in the same homology class. Turning on fluxes
generates a superpotential [9]

W = H ,
W(S) =
(3.5)
Sk + Nk Sk F0 .
k
In [6], in the case when all Nk s are positive, the N = 1 chiral superfield corresponding to Sk
is identified with the gaugino condensates of the SU(Nk ) subgroup of the U (Nk ) gauge group
1
factor, Sk = 32
2 TrSU(Nk ) W W , before the transition. When we have both branes and antibranes, as long as they are very far separated, this picture should persist. Namely, the brane
theory should still have gaugino condensation and confinement, and Sk s should still correspond
to the gaugino condensates, even though we expect the supersymmetry to be completely broken
in this metastable vacuum. Moreover, for each k, there is a remaining massless U (1) gauge field
in the dual geometry. It gets identified with the massless U (1) on the gauge theory side which
is left over by the gaugino condensation that confines the SU(Nk ) subgroup of the U (Nk ) gauge
theory.
In the supersymmetric case studied in [6], the coefficients of the polynomial f (x) determining
the dual geometry and the sizes of Si is fixed by the requirement that
Sk W (S) = 0
397
and this gives a supersymmetric holographic dual. In the case of interest for us, with mixed
fluxes, we do not expect to preserve supersymmetry. Instead we should consider the physical
potential V (S) and find the dual geometry by extremizing
Sk V (S) = 0
which we expect to lead to a metastable vacuum. The effective potential V is given in terms of
the special geometry data and the flux quanta:
V = g Si Sj Si WSj W,
where the Khler metric is given by gi j = Im(ij ) in terms of the period matrix of the CalabiYau
ij = Si Sj F0 .
In terms of , we can write

j k

2i

V=
(3.6)
j + jj N j k + kk N k ,

where we have all the j = . To be explicit, we consider in detail the case where we have only
two S 3 s in the dual geometry before turning to the more general case.
3.3. More precise statement of the conjecture
Let us make our conjecture precise: We conjecture that the large N limit of brane/anti-brane
systems are CalabiYaus with fluxes. In particular, the solutions to tree level string equations on
the closed string side lead to an all order summation of planar diagrams of the dual brane/antibrane system (i.e., to all orders in the t Hooft coupling). We translate this statement to mean that
the geometry of the closed string vacuum at tree level is captured by extremizing the physical
potential. Moreover the physical potential is characterized by the fluxes (which fix the superpotential) as well as by the Khler potential; the main ingredient for both of these objects is the
special geometry of the CalabiYau after transition. This, in turn, is completely fixed by tree
level topological string theory, or equivalently by the planar limit of a certain large N matrix
model [7]. Here we are using the fact that since N = 2 is softly broken by the flux terms, the special Khler metric is unaffected at the string tree level. It is quite gratifying to see that topological
objects, such as matrix integrals, play a role in determining the geometry of non-supersymmetric
string vacua!
Of course the Khler potential should be modified at higher string loops since. In particular
the 1/N corrections to our duality should involve such corrections. Note that, in the supersymmetric case studied in [6], there were no 1/N corrections to modify the geometry of the vacua.
We do expect the situation to be different in the non-supersymmetric case. Note however, from
the discussion of Section 2, the soft breaking is such that it is ambiguous which N = 1 supersymmetry the Lagrangian has. This should constrain what kind of quantum corrections one can
have beyond those allowed by a generic soft breaking. This deserves further study.
3.4. The case of 2 S 3 s
For simplicity we start with the case where we have just two S 3 s. Before the transition, there
are two shrinking P1 s at x = a1,2 . Let us denote by the distance between them,
= a 1 a2 .
398
The theory has different vacua depending on the number of branes we put on each P1 . The vacua
with different brane/anti-brane distributions are separated by energy barriers. To overcome these,
the branes must first become more massive.
The effective superpotential of the dual geometry, coming from the electric and magnetic FI
terms turned on by the fluxes, is
W(S) = (S1 + S2 ) + N1 S1 F0 + N2 S2 F0 .
The B-periods have been computed explicitly in [6]. We have

0
S1
+
S
)
log
2iS1 F0 = W (0 ) W (a1 ) + S1 log
2(S
1
2

g3
2
3
2
+ 2S1 10S1 S2 + 5S2 / g +
and

0
S2
1 2(S1 + S2 ) log
2iS2 F0 = W (0 ) W (a2 ) + S2 log

g3
2
3
2
5S1 10S1 S2 + 2S2 / g + ,
(3.7)
(3.8)
where the omitted terms are of order S1n2 S2n2 /(g3 )n1 +n2 1 , for n1 + n2 > 2. In the above,
0 is the cutoff used in computing the periods of the non-compact B cycles, and physically
corresponds to a high energy cutoff in the theory.
To the leading order (we will justify this a posteriori), we can drop the quadratic terms in the
Sk
s, and higher. To this order,
g3

2
S1
0
2i11 = 2iS21 F0 log
,
log
3

g
2
0
,
2i12 = 2iS1 S2 F0 log

2
S2
0
.
log
2i22 = 2iS22 F0 log

g3
In particular, note that at the leading order 12 is independent of the Si , so we can use ii as
variables. It follows easily that the minima of the potential are at
Re() + Re( )ij N j = 0
and

Im() + Im( )ij N j = 0.
For example, with branes on the first P1 and anti-branes on the second, N1 > 0 > N2 ,

S1 = g

S2 = g
0

0

2
2
0

0

2| N2 |
N1
2| N1 |
N2
e2i/|N1 | ,
e2i/|N2 | .
(3.9)
399
To see how to interpret this, let us recall the supersymmetric situation when N1 , N2 > 0. There,
to the same order, one finds
2(1+| N2 |)
2(1+| N1 |)
N1
N2
3 0
2i/|N1 |
3 0
S1 = g
e
,
S2 = g
e2i/|N2 | .

(3.10)
The interpretation of the above structure in the supersymmetric case is as follows. In the IR,
the theory flows to a product of two supersymmetric gauge theories with gauge group U (N1 )
U (N2 ), where each U (Ni ) factor is characterized by a scale i ,
S1 = 31 ,
S2 = 32 .
(3.11)
Let us denote by 1,2 the bare coupling constants of the low energy theory U (N1 ) and U (N2 )
theories, i.e.,
3|Ni |
i
2ii = log
(3.12)
, i = 1, 2.
0
On the one hand, we can simply read them off from (3.10) since we can write (3.11) as Si =
30 e2ii /|Ni | :

0 |N1 |
0 |N2 |
2i1 = 2i + log
2 log
,
m
mW

0 |N2 |
0 |N1 |
2i2 = 2i + log
(3.13)
2 log
.
m
mW
We could also have obtained this by using the matching relations with the high energy coupling . Namely, in flowing from high energies we have to take into account the massive fields
we are integrating out. The fields being integrated out in this case are the massive adjoint near
each vacuum, whose mass goes as m = g, and the massive W bosons in the bifundamental
representation, whose masses are mW = . The standard contribution of these heavy fields to
the running coupling constant is simply what is written in (3.13).
In the non-supersymmetric case, from (3.11) and (3.12), we can write the gaugino vevs in (3.9)
very suggestively exactly as in (3.13) except that we replace
|N1 |+|N2 |
|N1 |
|N |
|N | |N |
0 2 , 0 2 0 1 ,
and
W
mW m
leading to

0 |N1 |
0 |N2 |
2 log
,
2i1 = 2i + log
m
m
W

0 |N2 |
0 |N1 |
2 log
2i2 = 2i + log
m
m
W
(3.14)
which gives back (3.9). In other words, the leading result is as in the supersymmetric case, except
that branes of the opposite type lead to complex conjugate running. In particular, this implies
that the real part of the coupling runs as before, but the -angle is running differently due to the
400
matter field coming between open strings stretched between the branes and anti-branes. We will
discuss potential explanations of this in Section 3.6, where we discuss the matter structure in the
brane/anti-brane system.
The potential at the critical point is given by8

0 2
2
8
+

.
V = 2 |N1 | + |N2 | |N1 ||N2 | log
(3.15)
mW
gYM
The first term, in the holographic dual, corresponds to the tensions of the branes. The second
term should correspond to the ColemanWeinberg one loop potential which is generated by zero
point energies of the fields. This interpretation coincides nicely with the fact that this term is
proportional to |N1 ||N2 |, and thus comes entirely from the 12 sector of open strings with one
end on the branes and the other on the anti-branes. The fields in the 11 and 22 sectors with
both open string endpoints on the same type of brane do not contribute to this, as those sectors
are supersymmetric and the boson and fermion contributions cancel. We shall return to this in
Section 3.6. For comparison, in the case of where both P1 s were wrapped by D5 branes, the
potential at the critical point V++ equals
V++ =

8
|N1 | + |N2 | = V
2
gYM
and is the same as for all anti-branes. This comes as no surprise, since the tensions are the same,
and the interaction terms cancel since the theory is now truly supersymmetric.
We now consider the masses of bosons and fermions in the brane/anti-brane background.
With supersymmetry broken, there is no reason to expect the 4 real bosons of the theory, coming
from the fluctuations of S1,2 around the vacuum, to be pairwise degenerate. To compute them,
we simply expand the potential V to quadratic order. More precisely, to compute the physical
masses as opposed to the naive Hessian of the potential, we have to go to the basis where kinetic
terms of the fields are canonical. The computation is straightforward, if somewhat messy. At the
end of the day, we get the following expressions.

2 (a 2 + b2 + 2abcv) (a 2 + b2 + 2abcv)2 4a 2 b2 (1 v)2

,
m (c) =
2(1 v)2
where c takes values c = 1, and

N2
N1
,
,

b
a

3
3
21 Im 11
22 Im 22
(Im 12 )2
.
Im 11 Im 22
Indeed we find that our vacuum is metastable, because all the m2 > 0 (which follows from the
above formula and the fact that v < 1 in the regime of interest |Si /g3 | < 1). This is a nice
check on our holography conjecture, as the brane/anti-brane construction was clearly metastable.
Moreover, we see that there are four real bosons, whose masses are generically non-degenerate,
as expected for the spectrum with broken supersymmetry.
8 We have shifted the overall zero point energy of V as we did in the previous section, to make the brane and the
anti-brane tensions equal. More precisely, we set
8
V V + 2 (N1 + N2 ).
gYM
401
Since supersymmetry is completely broken from N = 2 to N = 0, we expect to find 2 massless Weyl fermions, which are the Goldstinos. More precisely, we expect this only at the closed
string tree level (i.e., at the leading order in 1/N expansion). Namely, at string tree level, we can
think about turning on the fluxes as simply giving them an expectation value, which would
break the N = 2 supersymmetry spontaneously. However, at higher orders, the Khler potential
should a priori not be protected in that breaking is soft, and not spontaneous. This will affect
the computation of the mass terms in Section 2, which relied on the N = 2 relation between
the Khler potential and the superpotential, and should result in only one massless fermion remaining. Masses of the fermions are computed from the N = 2 Lagrangian, as in Section 2, but
now with two vector multiplets. We again redefine the fields so as to give them canonical kinetic
terms. We indeed find two massless fermions, at the order at which we are working at. Since
supersymmetry is broken these are interpreted as the Goldstinos (we will give a more general
argument in the next subsection). There are also two massive fermions, with masses
mf1 =
a
,
1v
mf2 =
b
.
1v
Note that v controls the strength of supersymmetry breaking. In particular when v 0 the 4 boson masses become pairwise degenerate and agree with the two fermion masses a and b, as
expected for a pair of N = 1 chiral multiplets.
The mass splitting between bosons and fermions is a measure of the supersymmetry breaking.
In order for supersymmetry breaking to be weak, these splittings have to be small. There are two
natural ways to make supersymmetry breaking be small: One way is to take the number of antibranes to be much smaller than the number of branes. The other way is to make the branes and
anti-branes be very far from each other. We will consider mass splittings in both of these cases.
Adding a small anti-brane charge should be a small perturbation to a system of a large number of branes. In that context, the parameter that measures supersymmetry breaking should be
|N2 /N1 |, where |N2 | is the number of anti-branes. Let us see how this is reflected in v. We should
see that in this limit, v becomes small.
Note that (3.9) implies that

S1 |N1 | S2 |N2 | 2(|N1 |+|N2 |)
8 2

= 3
=
= exp (1 ) 2
g3

g
g
YM
with
gYM
2
2

0
|N1 | + |N2 | log .

Note that

log 0 /

=

1 log / .
(3.16)
2
is
For finite non-vanishing Si /g3 , clearly is very close to 1 because of the fact that gYM
very small for large cutoff 0 . Of course, this is compatible with the definition of as we recall
2 . Therefore, let us write
the running of gYM
= 1 .
402
Fig. 5. Here we depict the masses of the four bosons as solid lines and the two non-vanishing fermionic masses as
dashed lines. The other two fermions, not shown here, are massless Goldstinos. Note that supersymmetry breaking is
most pronounced when |N1 | = |N2 |. This plot is for fixed |N1 N2 |, as we vary |N1 /N2 |.
We then have (without any approximation)

v=
(1 )2
(Im 12 )2
=
.
Im 11 Im 22 (1 + 2 ) + (| N2 | + | N1 |)
N1
N2
(3.17)
|N2 |
1|
And in the limit of |N
|N2 | (or similarly when |N1 | ) we find that v 0. This is as
expected, since in this limit we expect supersymmetry breaking to be very small.
In Fig. 5 are plots of the masses for non-supersymmetric vacua, with fixed |N1 N2 | but varying
2
log | N
N1 |. They demonstrate how mass splitting vanishes in the extreme limits of the ratio N1 /N2 ,
where supersymmetry is restored.
It is also natural to consider the mass splittings in the limit of large separation of branes, where
supersymmetry is expected to be restored. We have to be somewhat careful in taking this limit.
It turns out that the right limit is when
|| || |0 |
and where
|/| |0 /|
which implies from (3.16) that 0 and from (3.17) we have v 0 and thus the mass splittings
disappear as expected.
It is also interesting to write the energy of the vacuum (3.15) as a function of v. In particular let
E represent the shift in energy from when the branes are infinitely far away from one another.
Then we find from (3.15) that
E

.
=
|N2 | 2
E
1|
( |N
+
)
|N2 |
|N1 |
Note that the energy shift goes to zero as |N2 /N1 | 0, as expected. It also goes to zero in the
large brane/anti-brane separation, because then 0. The sign of the shift is also correct, as one
403
would expect that the attraction between the branes and the anti-branes should would decrease
the energy of the system. Note also that for fixed |N1 N2 |, E/E, as well as the mass splittings, is
maximal when N1 = N2 , i.e., when we have equal number of branes and anti-branes, whereas
if one of them is much larger than the other, mass splittings as well as E/E 0. This is again
consistent with the fact that in the limit with an extreme population imbalance of branes, the
supersymmetry breaking goes away.
3.5. n S 3 s
Let us now generalize our analysis to the case with an arbitrary number of blown up S 3 s.
In this case, before the geometric transition there are n shrinking P1 s located at x = ai , i =
1, 2, . . . , k. We will denote the distance between them as
ij = ai aj .
As in the simpler case, for a specified total number of branes, we can classify the vacua of the
theory by the distribution of the branes among the P1 s, and these vacua will be separated by
potential barriers corresponding to the increase in the size of the wrapped P1 between critical
points.
The effective superpotential of the dual geometry after the geometric transition, in which the
branes have been replaced by fluxes, is given by (3.4). The B-periods will now take the form
2
ij
Si
+
Sj log
+ ,
2iSi F0 = Si log
2

W (ai )0 j =i
20
where again, 0 is interpreted as a UV cutoff for the theory, and we are omitting terms analogous
to the polynomial terms in the example of the last section.
To leading order, the generalization of the 2 S 3 case to the current situation is straightforward.
We have

S1
2iii = 2iS2i F0 log
,
W (ai )20
2
0
.
2iij = 2iSi Sj F0 log
2ij
The potential (3.7) has its critical points where

i

Si V = Fij l g Sj Sj g Sk Sk j + jj N j k + kk N k = 0.
2
The solutions which correspond to minima are then determined by
(3.18)
Re() + Re( )ij N j = 0

and

Im() + Im( )ij N j = 0.
Indeed, in the case where all N i are either positive (negative), these are just the (non-)standard
F-flatness conditions of the N = 1 supersymmetric theory, so they will be satisfied exactly. In
the non-supersymmetric cases, the conditions will receive perturbative corrections coming from
the corrections to the Khler potential at higher string loops.
404
The expectation values of the bosons S i take a natural form in these vacua, being given by
Si = W
Nj >0
(ai )20
j =i
Si = W
Nj >0
(ai )20
j =i
0
ij
0
ij
2 |Nj | N
|Nk |
k <0
|Ni |
0 2 |Ni |
k=i
ik
2 |Nj | N
|Nk |
k <0
|Ni |
0 2 |Ni |
k=i
ik

2i
exp
,
|Ni |

2i
exp
,
|Ni |
Ni > 0,
Ni < 0.
The explanation of these expectation values is the natural generalization of the case from the
previous section. We characterize each gauge theory in the IR by a scale i , where Si = 3i .
The massive adjoints near the ith critical point now have mi = W (ai ), whereas the massive W
bosons have mass mWij = ij . Then matching scales, we find

Sibrane
|Ni |
|N |
Sianti-brane i
3|Ni |
= i
= e2i 0
3|N |
= i i
2Nbrane
=e
2i
|N |
2N
0 anti-brane mii
same

j =i
|N |
2N
2N
0 brane 0 anti-brane mii
2Nj
same

j =i
opposite
mWij
i
m2N
Wij
mWij 2Nj ,
j =i
opposite
mWij 2Nj .
j =i
Above, Nbrane and Nanti-brane are the total number of branes and anti-branes, respectively, and
the products are over the sets of branes of the same and opposite types as the brane in question. In the supersymmetric case, this reduces to the expected relation between the scales. The
non-supersymmetric case is, to leading order, identical except that the branes of opposite type
contribute complex conjugate running, which is explained in Section 3.6.
Similarly to the 2 S 3 case, the gauge coupling constant in each U (Nk ) factor will run as

opposite
same
0 |Nj |
0 |Ni |
0 |N2 |
2
log
2
log
,
2ii = 2i + log
mi
mWij
m
Wij
j =i
j =i
for a brane (Ni > 0). For anti-brane (Ni < 0) and the running is

opposite
same
0 |Nj |
0 |Ni |
0 |N2 |
2ii = 2i + log
2
log
2
log
.
mi
mWij
m
Wij
j =i
j =i
This again has a simple form corresponding to integrating out massive adjoints and W bosons.
Moreover, the fields corresponding to strings with both endpoints on branes, or on the anti-branes,
contribute to running of the gauge coupling as in the supersymmetric case. The fields coming
from strings with one end on the brane and the other on the anti-brane, again give remarkably
simple contributions, almost as in the supersymmetric case: the only apparent difference being
in how they couple to the angle. We will come back to this later.
One could in principle compute the masses of the bosons and fermions in a generic one of
these vacua. We will not repeat this here. Instead, let us just try to understand the massless
fermions, or the structure of supersymmetry breaking in this theory. The results of Section 2 are
405
easy to generalize, to find the fermion mass matrices9

i

m i j = g Sl Sk Fij l k + kk N k ,
2

i Sl Sk

Fij l k + kk N k .
mi j = g
2
In the above, the repeated indices are summed over, and we have denoted
Fij k =
3 F0
.
Si Sj Sk
At the order to which we are working (corresponding to one loop on the field theory side), Fij k
is diagonal. The masses then imply that in vacua where the standard F-term
Si W = i + ij N j
fails to vanish, a zero eigenvalue is generated for the mass matrix, while when the non-standard
F-term
= i + ij N j
Si W
fails to vanish, a zero eigenvalue is generated for the mass matrix.
More generally, note that the exact conditions (at leading order in 1/N ) for critical points of
the potential (3.18) can be re-written in terms of the fermion mass matrices as

m i j g Sj Sk k + kk N k = 0,
(3.19)

mi j g Sj Sk k + kk N k = 0.
(3.20)
In the case when supersymmetry is unbroken, and all the P1 s are wrapped by branes, we will
find both of these equations to be truly exact, and satisfied trivially: the mass matrix vanish
identically, so (3.20) is trivial, and (3.19) is satisfied without constraining m i j directly since

the vanishing of (k + kk N k ) is the F-flatness condition. A similar story holds in the anti-brane
vacuum.
In the case with both branes and anti-branes, the equations are not trivial. Instead, they say
that both the - and the -mass matrices have at least one zero eigenvalue. At the string tree
level (i.e., at the leading order in 1/N expansion), we expect them to have exactly one zero
eigenvalue each. To this order, we can think about turning on the fluxes as simply giving them
an expectation value, which would break the N = 2 supersymmetry spontaneously. However,
at higher orders, the Khler potential is not protected in that the symmetry breaking is soft,
but not spontaneous. Consequently, the form of Eqs. (3.19), (3.20) should presumably receive
corrections, and only one massless fermion should remain, corresponding to breaking N = 1 to
N = 0.
The leading-order potential in these vacua is given by

Ni >0,N
j <0 2
0
8
.
|Ni |
|Ni ||Nj | log
V =

g 2
YM
i,j
ij
9 Note that the mass matrices are now being expressed not in the basis in which the kinetic terms take the canonical
form, but in the basis in which the Lagrangian was originally expressed. This will not affect our analysis.
406
This takes the natural form of a brane-tension contribution for each brane in the system, plus
interaction terms between the brane/anti-brane pairs. If we define

0
gYM 2
.
ij
|Nk | log
2
ij
k
Then in terms of these parameters ij , the physical potential takes the simpler form of
Ni >0,Nj <0

ij |Ni ||Nj |
8
i,j

V =
|N
|
i
gYM 2
i |Ni |
i
so that the energy shift is

E
=
E
Ni >0,Nj <0

i,j
ij |Ni ||Nj |

.
( i |Ni |)2
Once again, the forces among the (anti-)branes cancel. Moreover, the branes and the anti-branes
attract, which lowers the energy of the vacuum, below the supersymmetric one. This energy shift
will effectively vanish only when there is an extreme imbalance between number of the branes
and anti-branes, just as in the 2 S 3 case.
3.6. Brane/anti-brane gauge system
It is natural to ask whether we can find a simple description of the non-supersymmetric theory
corresponding to the brane/anti-brane system. This certainly exists in the full open string field
theory. However, one may wonder whether one can construct a field theory version of this which
maintains only a finite number of degrees of freedom. In fact, it is not clear if this should be
possible, as we will now explain.
Consider a simpler situation where we have a stack of N parallel D3 branes. This gives an
N = 4 supersymmetric U (N ) gauge theory. Now consider instead (N + k) D3 branes and k antiD3 branes separated by a distance a. This is clearly unstable, and will decay back to a system
with N D3 branes. Can one describe this in a field theory setup? There have been attempts
[18] along these lines (in connection with Sens conjecture [19]), however no complete finite
truncation seems to exist. This system is similar to ours. In particular, consider the case with 2
S 3 s, with a total of N D5 branes. We know that we can have various supersymmetric vacua of
this theory in which
U (N ) U (N1 ) U (N N1 ),
where we have N1 branes wrap one P1 and N N1 wrap the other. However we also know that
the theory with N D5 branes can come from a meta-stable vacuum with (N + k) D5 branes wrap
around one P1 and k anti-D5 branes around the other. This theory would have a U (N + k) U (k)
gauge symmetry. Clearly we need more degrees of freedom than are present in the U (N ) theory
in order to describe this theory. In fact, since k can be arbitrarily large, we cannot have a single
system with a finite number of degrees of freedom describing all such metastable critical points.
Of course, one could imagine that there could be a theory with a finite number of fields which
describes all such configurations up to a given maximum k, but this does not seem very natural. It
is thus reasonable to expect that string field theory would be needed to fully describe this system.
407
Nevertheless we have seen, to leading order, a very simple running of the two gauge groups
in the non-supersymmetric case, and it would be natural to ask if we can explain this from the
dual open string theory. In this dual theory, where we have some branes wrapping the first P1
and some anti-branes the other, we have three different types of sectors: the 11, 22 and the
12 open string sectors. The 11 and the 22 subsectors are supersymmetric and give rise to
the description of the N = 1 supersymmetric U (Ni ) theories for i = 1, 2, coupled to the adjoint
matter fields which are massive. This part can be inherited from the supersymmetric case,
and in particular explains the fact that the running of the coupling constant of the U (Ni ) from
the massive field is the same as in the supersymmetric case. The difference between our case
and the supersymmetric case comes from the 12 subsector. In the NS sector we have the usual
tachyon mode whose mass squared is shifted to a positive value, as long as is sufficiently larger
than the string scale, due to the stretching of the open string between the far separated P1 s. We
also have the usual oscillator modes. In the Ramond sector ground state the only difference
between the supersymmetric case and our case is that the fermions, which in the supersymmetric
case is the gaugino partner of the massive W12 boson, has the opposite chirality. The fact that
they have the opposite chirality explains the fact that the term of the gauge group runs with
an opposite sign compared to the supersymmetric case. In explaining the fact that the norm runs
the same way as the supersymmetric case, the contributions of the fermions is clear, but it is not
obvious why the NS sector contribution should have led to the same kind of running. It would
be interesting to explain this directly from the open string theory annulus computation in the NS
sector.
It is easy to generalize the discussion above to the n S 3 case. In fact just as in the 2 S 3 case,
the only subsectors which are non-supersymmetric are the ij sectors where i is from a brane
and j is from an anti-brane cut. This agrees with the results obtained in the general case, where
supersymmetry breaking contributions come, to leading order, precisely from these sectors.
4. Decay rates
In the previous sections, we constructed non-supersymmetric vacua of string theory corresponding to wrapping D5 branes and anti-D5 branes on the 2-cycles of the local CalabiYau.
Here, branes and anti-branes are wrapping rigid 2-cycles which are in the same homology class,
but are widely separated, and there are potential barriers between them of tunable height. As
we emphasized, the crucial aspect of this is that the vacua obtained in this way can be long
lived, and thus are different from braneanti-brane systems in flat space, or systems of branes
and anti-branes probing the CalabiYau manifold, which have been considered in the literature.
Moreover, when the numbers of branes on each 2-cycle is large, this has a holographic dual in
terms of non-supersymmetric flux vacua. By duality, we also expect the corresponding flux vacua
to be metastable, despite breaking supersymmetry. In this section we will explore the stability of
these vacua in more detail.
In both open and closed strings language, the theory starts out in a non-supersymmetric vacuum. Since there are lower energy states available, the non-supersymmetric vacuum is a false
vacuum. As long as the theory is weakly coupled throughout the process, as is the case here, the
decay can be understood in the semi-classical approximation. This does not depend on the details
of the theory, and we will review some aspects of the beautiful analysis in [20]. Afterwards, we
apply this to the case at hand.
408
The false vacuum decays by nucleating a bubble of true vacuum by an instanton process. The
rate of decay is given in terms of the action of the relevant instanton as
exp(SI ).
The instanton action SI is the action of the Euclidean bounce solution, which interpolates between the true vacuum inside the bubble and the false vacuum outside of the bubble.10 Assuming
that the dominant instanton is a spherical bubble of radius R, the Euclidean instanton action is
given by [20]
2 4
R V + 2 2 R 3 SD .
2
The first term is the contribution to the action from inside the bubble. Here V is the difference
in the energy density between the true and the false vacuum, and this is multiplied by the volume
of the bubble (a 4-sphere of radius R). The second term is the contribution to the action from
the domain wall that interpolates between the inside and the outside on the bubble, assuming the
domain wall is thin. There, SD is the tension of the domain wall, and 2 2 R 3 is its surface area
(the area of a 3-sphere of radius R). The radius R of the bubble is determined by energetics: The
bubble can form when the gain in energy compensates for the mass of the domain wall that is
created. The energy cost to create a bubble of radius R is
SI =
4 3
R V + 4R 2 SD ,
3
where the first term comes from lowering the energy of the vacuum inside the bubble, and the
last term is the energy cost due to surface tension of the domain wall. We can create a bubble at
no energy cost of radius
E=
SD
.
V
Bubbles created with radio smaller than this recollapse. The bubbles created at R = R expand
indefinitely, as this is energetically favored. At any rate, the relevant instanton action is SI (R ),
or
R = 3
SI =
4
27 SD
.
2 (V )3
Correspondingly, the vacuum is longer lived the higher the tension of the domain wall needed to
create it, and the lower the energy splitting between the true and the false vacuum.
With this in hand, let us consider the decays of the vacua we found. When the number of
branes wrapping each P1 is small, the open string picture is appropriate. Consider, for simplicity,
the case where the superpotential has only two critical points
W (x) = g(x a1 )(x a2 ),
and we have only a single brane at x = a1 and a single anti-brane at x = a2 . The effective
potential that either of these branes sees separately is given simply by the size of the 2-sphere it
10 The coefficient of proportionality comes from the one loop amplitude in the instanton background, and as long as S
I
is large, its actual value does not matter.
wraps
2
1/2

,
A(x) = W (x) + |r|2
409
(4.1)
(see Section 3). The branes are,

where |r| is the tension of the brane wrapping the minimal
of course, also charged, so there is an electric field flux tube (of the six-form potential) running
between the brane and the anti-brane, along a 3-cycle corresponding to an S 2 being swept from
x = a1 to x = a2 . Despite the flux tube, the system is (meta)stable if the potential barrier from
(4.1) is high enough. The decay process corresponds, for example, to having the D5 brane tunnel
under the energy barrier from the vacuum at x = a1 to the vacuum x = a2 , to annihilate with the
anti-D5 brane there. More precisely, what happens is that a bubble of true vacuum is created in
R 3,1 , inside of which the branes are annihilated. The boundary of the bubble is a domain wall
which, in the thin wall approximation, corresponds to a D5 brane wrapping the S 2 boundary of
the bubble in R 3 and the 3-cycle in the CalabiYau where the flux tube was. The tension of the
domain wall is the size of the 3-cycle that the D5 brane wraps in the CalabiYau times the tension
of the 5-brane
a2

1
1
1 3
A(x) dx W (a2 ) W (a1 ) =
g ,
SD =
gs
gs
3gs
P1
a1
where the second term in (4.1) gives vanishingly small contribution11 in the limit r. The
difference in the cosmological constants of the two vacua corresponds, to the leading order, to
the difference in tensions between the original configuration, with Vi = 2|r|/gs and the final one
where the branes have annihilated Vf = 0, so the gain in energy is simply
V = 2
2
|r|
= 2 .
gs
gYM
When the number of branes wrapping the P1 s is large, we use the dual geometry from after
the transition, with fluxes. Consider, for example, the case when N1 branes wraps the first P1 and
|N2 | anti-branes the second, with |N1,2 | large, and satisfying
N1 > 0 > N 2 .
Then, the C cycle that runs between the two cuts, [C] = [B1 ] [B2 ] is a compact flux tube, with
|N2 | units of H RR flux running through it. To decay, one kills off the flux lines one by one, by
creating a D5 brane wrapping a zero size P1 where the first P1 used to be before the transition.
This brane grows and eats up the flux line until it vanishes again at the first cut. After all the flux
has decayed, this leaves us with N1 + N2 units of flux through the first S 3 . The tension of the
corresponding domain wall is

|N2 |
||,
SD =
gs
C
where

=
B1
W (a2 ) W (a1 ),
B2
11 For us, the mass of the brane comes from the B-field only, and the brane is heavy only because the string coupling is
weak.
410
where we omit terms that are exponentially suppressed in the vacuum.12 In this case, the energy
of the false vacuum is, to the leading order, Vi = (|N1 | + |N2 |)|r|/gs . After the flux has decayed,
we have Vf = (|N1 | |N2 |)|r|/gs coming from the flux through first S 3 only. The difference is
just
V = Vi Vf 2|N2 |
|r| 2|N2 |
= 2 .
gs
gYM
We see that both the domain wall tension and the difference in vacuum energies before and after
are equal to what we found in the open string picture, albeit rescaled by the amount of brane
charge which disappears! This is, of course, not surprising because the domain wall relevant in
the closed string case is the same one as that relevant in the open string case, just before the large
N transition.
We can now put everything together to compute the action of the instanton corresponding to
nucleating a bubble of true vacuum:
SI
|N2 | |(g3 )4 |
.
48 gs
|r|3
This leads to the decay rate

exp(SI ).
Note that the relevant instantons come from D-branes wrapping cycles, and thus it is natural for
the instanton action to depend on gs in the way the D-brane action does.
This exactly reproduces the results one would have expected from our discussion in Section 3.
There are two effects controlling the stability of the system. Namely, one is the height of the
potential barrier, which is controlled by here by g3 and which enters the action of the domain
wall. The branes and anti-branes become more separated the larger the , and their interactions
weaker. In addition, increasing the height of the potential barrier by making larger, overcomes
the electrostatic attraction of the branes. The other relevant parameter is the relative difference
in energies of the true and the false vacuum, which depends on the ratio of brane to anti-brane
numbers |N2 /N1 |. When this is very small, the vacuum with both the branes and the anti-branes
is nearly degenerate with the vacuum with just the branes, and correspondingly, the decay time
gets longer as V /V gets smaller.
5. Open questions
In this paper, we have seen how a large N system of branes which geometrically realizes
metastability by wrapping cycles of a non-compact CY can be dual to a flux compactification
which breaks supersymmetry. We provided evidence for this duality by studying the limit where
the cycles of the CalabiYau are far separated.
There is a number of open questions which remain to be studied: It would be nice to write
down a non-supersymmetric gauge theory with finite number of degrees of freedom which describes this brane configuration. It is not clear that this should be possible, but it would be
12 More precisely, we should evaluate the corrections to this at the values of S
1,2 corresponding to being somewhere in
1,2
, and
the middle between the true and the false vacuum. Either way, the omitted terms are suppressed by, at least,
we can neglect them.
411
interesting to settle this question conclusively one way or the other. This is also important for
other applications in string theory where dynamics of brane/anti-brane systems is relevant.
Another issue which would be important to study is the phase structure of our system. Since
we have broken supersymmetry, we have no holomorphic control over the geometry of the solutions. It would be interesting to see what replaces this, and how one should think of the global
phase structure of these non-supersymmetric solutions. Since the full potential is characterized
by the N = 2 prepotential and the flux data, which in turn are characterized by integrable data
(matrix model integrals of the superpotential), one still expects that the critical points of the
potential should be characterized in a nice way. Understanding this structure would be very important. For example, it would tell us what happens when the separation of the wrapped cycles is
small where one expects to lose metastability due to the open string tachyons stretching between
branes and anti-branes.
One other issue involves the 1/N corrections to the holographic dual theory. In this paper we have focused on the leading order in the 1/N expansion, i.e., at the string tree level.
How about subleading corrections? This is likely to be difficult to address, as one expects nonsupersymmetric corrections to string tree level to be difficult to compute. The fact that we have
a metastable system suggests that we would not want to push this question to an exact computation, as we know the system is ultimately going to decay to a stable system. However, we would
potentially be interested in studying the regions of phase space where metastability is just being
lost, such as when the cycles are close to one another.
Finally, perhaps the most important question is about the embedding of our mechanism for
inducing metastable flux compactification vacua into a compact CalabiYau geometry (see [21]).
At first sight, it may be unclear whether there would be any obstacles to a compact embedding
of our story. There is at least an example similar to what we are expecting in the compact case.
In [22], among the solutions studied, it was found that flipping the signs of some fluxes leads to
apparently metastable non-supersymmetric vacua. It would be interesting to see if this connects
to the mechanism introduced in this paper, where the existence of metastability for flux vacua
is a priori expected. Moreover, from our results it is natural to expect that most flux vacua do
have metastable non-supersymmetric solutions. For instance, in the non-compact case we have
studied, only 2 out of 2n choice of signs for the n fluxes 2 were non-supersymmetric. This
suggests that in the study of flux compactifications, the most natural way to break supersymmetry
is simply studying metastable vacua of that theory, without the necessity of introducing any
additional anti-branes into the system. It would be very interesting to study this further.
Acknowledgements
We would like to thank N. Arkani-Hamed, K. Becker, M. Becker, R. Bousso, E. Imeroni,
K. Intriligator, K. Kim, J. McGreevy, L. Motl, D. Sahakyan, D. Shih, W. Taylor and B. Zwiebach
for valuable discussions. We would also like to thank the Stony Brook Physics department and
the fourth Simons Workshop in Mathematics and Physics for their hospitality while this project
was initiated.
The research of M.A. and C.B. is supported in part by the UC Berkeley Center for Theoretical Physics. The research of M.A. is also supported by a DOI OJI Award, the Alfred P. Sloan
Fellowship and the NSF grant PHY-0457317. The research of J.S. and C.V. is supported in part
by NSF grants PHY-0244821 and DMS-0244464. The research of J.S. is also supported in part
by the Korea Foundation for Advanced Studies.
412
References
[1] M.R. Douglas, S. Kachru, Flux compactification, hep-th/0610102.
[2] K. Intriligator, N. Seiberg, D. Shih, Dynamical SUSY breaking in meta-stable vacua, JHEP 0604 (2006) 021, hepth/0602239.
[3] H. Ooguri, Y. Ookouchi, Meta-stable supersymmetry breaking vacua on intersecting branes, hep-th/0607183;
S. Franco, I. Garcia-Etxebarria, A.M. Uranga, Non-supersymmetric meta-stable vacua from brane configurations,
hep-th/0607218;
I. Bena, E. Gorbatov, S. Hellerman, N. Seiberg, D. Shih, A note on (meta)stable brane configurations in MQCD,
hep-th/0608157.
[4] S. Kachru, J. McGreevy, Supersymmetric three-cycles and (super)symmetry breaking, Phys. Rev. D 61 (2000)
026001, hep-th/9908135;
K.A. Intriligator, N. Seiberg, S.H. Shenker, Proposal for a simple model of dynamical SUSY breaking, Phys. Lett.
B 342 (1995) 152, hep-ph/9410203.
[5] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[6] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001) 3,
hep-th/0103067.
[7] R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[8] G. Veneziano, S. Yankielowicz, An effective Lagrangian for the pure N = 1 supersymmetric YangMills theory,
Phys. Lett. B 113 (1982) 231.
[9] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69, hep-th/9906070;
S. Gukov, C. Vafa, E. Witten, Nucl. Phys. B 608 (2001) 477, Erratum.
[10] T.R. Taylor, C. Vafa, RR flux on CalabiYau and partial supersymmetry breaking, Phys. Lett. B 474 (2000) 130,
hep-th/9912152.
[11] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: Duality cascades and chiSB-resolution of
naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[12] J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super-YangMills, Phys. Rev. Lett. 86 (2001)
588, hep-th/0008001.
[13] I. Antoniadis, H. Partouche, T.R. Taylor, Spontaneous breaking of N = 2 global supersymmetry, CPTH-S397.1195
LPTENS-95/55, 2006, hep-th/9512006.
[14] E.A. Ivanov, B.M. Zupnik, Modified N = 2 supersymmetry and FayetIliopoulos terms, JINR E2-97-322, 1997,
hep-th/9710236.
[15] J.D. Lykken, Introduction to supersymmetry, hep-th/9612114.
[16] M. de Roo, J.W. van Holten, B. de Wit, A. Van Proeyen, Chiral superfields in N = 2 supergravity, Nucl. Phys. B 173
(1980) 175.
[17] S. Kachru, S. Katz, A.E. Lawrence, J. McGreevy, Open string instantons and superpotentials, Phys. Rev. D 62
(2000) 026001, hep-th/9912151;
M. Aganagic, C. Vafa, Mirror symmetry, D-branes and counting holomorphic discs, hep-th/0012041;
F. Cachazo, S. Katz, C. Vafa, Geometric transitions and N = 1 quiver theories, hep-th/0108120.
[18] J.A. Minahan, B. Zwiebach, Gauge fields and fermions in tachyon effective field theories, JHEP 0102 (2001) 034,
hep-th/0011226;
J.A. Minahan, B. Zwiebach, Effective tachyon dynamics in superstring theory, JHEP 0103 (2001) 038, hepth/0009246;
J.A. Minahan, B. Zwiebach, Field theory models for tachyon and gauge field string dynamics, JHEP 0009 (2000)
029, hep-th/0008231.
[19] A. Sen, Tachyon condensation on the brane anti-brane system, JHEP 9808 (1998) 012, hep-th/9805170.
[20] S.R. Coleman, The fate of the false vacuum. 1. Semiclassical theory, Phys. Rev. D 15 (1977) 2929;
S.R. Coleman, Phys. Rev. D 16 (1977) 1248, Erratum;
C.G. Callan, S.R. Coleman, The fate of the false vacuum. 2. First quantum corrections, Phys. Rev. D 16 (1977)
1762.
[21] T. Banks, M. Dine, L. Motl, On anthropic solutions of the cosmological constant problem, JHEP 0101 (2001) 031,
hep-th/0007206.
[22] O. DeWolfe, A. Giryavets, S. Kachru, W. Taylor, Type IIA moduli stabilization, JHEP 0507 (2005) 066, hepth/0505160.
Nuclear Physics B 789 [FS] (2008) 413451
Strong coupling limit of Bethe ansatz equations

Ivan Kostov a,b,,1 , Didina Serban a,b , Dmytro Volin a,c
a Service de Physique Thorique, CNRS-URA 2306, C.E.A.-Saclay, F-91191 Gif-sur-Yvette, France
b Kavli Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA
c Bogolyubov Institute for Theoretical Physics, 14b Metrolohichna Str., Kyiv 03143, Ukraine
Received 7 May 2007; accepted 19 June 2007

Available online 28 June 2007
Abstract
We develop a method to analyze the strong coupling limit of the Bethe ansatz equations supposed to
give the spectrum of anomalous dimensions of the planar N = 4 gauge theory. This method is particularly
adapted for the three rank-one sectors, su(2), su(1|1) and sl(2). We use the elliptic parametrization of the
Bethe ansatz variables, which degenerates to a hyperbolic one in the strong coupling limit. We analyze the
equations for the highest excited states in the su(2) and su(1|1) sectors and for the state corresponding to
the twist-two operator in the sl(2) sector, both without and with the dressing kernel. In some cases we were
able to give analytic expressions for the leading order magnon densities. Our method reproduces all existing
analytical and numerical results for these states at the leading order.
1. Introduction
There is mounting evidence that the planar N = 4 SYM gauge theory is integrable [13] to
higher orders in perturbation theory, nourishing the hope to prove by this mean its equivalence
with the string theory on AdS5 S 5 [46]. Under the assumption of integrability, the spectrum
of the anomalous dimensions of the planar N = 4 SYM gauge theory is encoded, at least for
states with large number of operators, in the Bethe ansatz equations. The S-matrix behind these
* Corresponding author at: Service de Physique Thorique, CNRS-URA 2306, C.E.A.-Saclay, F-91191 Gif-sur-Yvette,
France.
E-mail address: kostov@spht.saclay.cea.fr (I. Kostov).
1 Associate member of the Institute for Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 72
Tsarigradsko Chausse, 1784 Sofia, Bulgaria.
doi:10.1016/j.nuclphysb.2007.06.017
414
I. Kostov et al. / Nuclear Physics B 789 [FS] (2008) 413451
equations [7] is fixed by symmetry [8,9], up to a scalar factor which is supposed to obey a crossing symmetry relation [10]. At strong coupling, the scalar factor should allow to reproduce the
string theory results. Its first two orders in the strong coupling expansion were fixed in this way
[11,12]. More recently, based on previous work of Beisert et al. [13,14] proposed an expression
for the scalar factor which is crossing symmetric and reproduces both the strong coupling, string
results and the perturbative results for the anomalous dimensions in the gauge theory. An efficient way of testing the predictions of the Bethe ansatz equations proved to be via the anomalous
dimension of the twist-two operator. In the regime of large Lorentz spin S, this quantity scales
logarithmically
S = f (g) ln S + .
(1.1)
The universal scaling function for N = 4 SYM up to third loop order was extracted [15] from
a perturbative three-loop computation in QCD [16]. The same quantity appears in the iterative
structure [17] of the multi-gluon amplitudes of the supersymmetric gauge theory. It was computed up to three loop order in [18], and, by an impressing effort, up to four loop order in [19,20].
The four loop result agrees with the prediction based on the proposal [14]. On the other hand, at
strong coupling the universal scaling function f (g) is predicted to behave as [21,22]
3 ln 2
+ .
Here we adopt the recent convention for the coupling constant g

f (g) = 4g
(1.2)
.
(1.3)
16 2
After the proposal [14] was made, several groups [2325] attempted to derive from it the strong
coupling expression (1.2). While the numerical work [24] easily reproduced the behavior (1.2)
and even predicted the next term in the expansion, the analytical treatment proved to be much
more difficult. Up to now, only the first term was obtained analytically [23,25].
One of the aspects which deserve further attention is to understand the origin of the dressing
phase [14] and to test its validity. It is believed that the dressing phase comes from a non-trivial
structure of the vacuum. Very recently, a structure similar to that of the dressing phase was
obtained [26], via the nested Bethe ansatz, for one of the non-trivial vacuum states in one of
the sectors which are not of rank one.
The aim of the present paper is to develop a systematic approach for solving the Bethe ansatz
equations at strong coupling. Our analysis is based on the formulation of the Bethe equations
as integral equations, and it works most simply on, although it is not restricted to, states as the
antiferromagnetic state in the su(2) sector, on the most excited state in the su(1|1) sector, or for
the finite-twist operators, in the sl(2) sector. Some of this cases were already studied numerically
or analytically; here we aim to a unitary treatment and hope that this method will be useful to
obtain systematical 1/g expansion.
Our starting point is the elliptic parametrization of the variables u and x that appear in
the Bethe equations. The elliptic parametrization we use here is related to the one proposed by
Janik [10] by a GaussLanden transformation, and it appeared independently in [9] and [27]. The
elliptic modulus is defined by
g2 =
k
1
=
.
k
4g
(1.4)
415
In the hyperbolic limit g , this parametrization supplies natural variables in the different
regimes, recently discussed in [13,28].
As it became clear from the numerical solutions of the integral equations [29], the points u =
2g become singular for large g and they split the real axis u into two regions, with |u|/2g less
or larger than one, where two different hyperbolic parametrizations apply.2 The region |u| < 2g
is that of the giant magnons, in the language of [28], and is characterized by finite values of
the periodic momenta p. The region |u| > 2g corresponds to the plane-wave limit [32], where
the momenta are of the order p 1/g.
It is more difficult to see what happens exactly at the points u = 2g and in their vicinity.
These points separate the two regimes discussed above, and they are not properly described by
either of the two parametrizations. In fact, they correspond to momenta of the order p g 1/2 ,
or to the near-flat space region, again in the terminology of [28]. In some cases almost all
the roots of the Bethe equations are concentrated in this region. It is possible to use the elliptic
parametrization to obtain the strong coupling expansion in this regime,3 and it is clear that the
expansion in this region involves powers of g 1/2 , or equivalently 1/1/4 .
In order to obtain the strong coupling limit of the dressing kernel [14] we have several options. One is to take the large coupling limit term by term in the series defining the dressing
phase around g = [13,14]. For the giant magnon and plane-wave limits this works, and this is
probably the most straightforward way to obtained the leading orders. In the near-flat space limit,
however, as it was noticed in [28], all the terms of the series contribute to the leading order in
1/g. We would like to emphasize that, although the results for the leading order of the anomalous
dimensions can be obtained without being particularly careful about the near-plane wave regime,
the corrections will be crucially determined by it.
A second possibility is to use the representation of the dressing kernel given by the magic
formula of [14], and this representation was already exploited in [25]. In this paper we give
another related integral representation
Kd (u, u ) = 4

n,1

1,n
n,1
dv K
(u, v) K+
(v, u ) K+
(v, u ) ,
(1.5)
n1
where the kernels K (u, u ) are the odd/even parts of the basic su(1|1) kernel K(u, u ), the
inverse Fourier transform counterparts of the kernels K0,1 (t, t ) in the BES decomposition [14].
m,n
The upper indices K
(u, u ) mean essentially that the variables u and u are defined by elliptic
functions of modulus
n
kn
=
,
kn 4g

m
km
=
.
km 4g
(1.6)
Similar variables were used in [13] and they are associated to bound states of magnons [35,36].
The sum in (1.5) resembles to a sum over intermediate states; in particular, the energy associated
to these intermediate states is of the type [35]

En = n2 + 16g 2 sin2 p/2.
(1.7)
2 The two hyperbolic parametrizations are related by the transformation s K s, where K is the real period of the
elliptic function.
3 In this case, the parametrization is obtained by shifting s K/2 + s.
416
In the large g limit, the sum over n can be taken with the help of the EulerMaclaurin formula
and the integrals over the variables v and z = n/4g are simple enough to be done. After taking the
integrals, we reproduce the first term in the strong coupling expansion of the dressing kernel [11].
Although technically this result is not very useful, it can be considered as another evidence that
the weak coupling and strong coupling definitions of the dressing kernel [14] agree.
Turning to the explicit results, we obtain here the leading order result for anomalous dimensions of the highest excited state in the su(2) and su(1|1) sectors, both for the Bethe ansatz with
the Beisert, Hernndez, Lpez/Beisert, Eden, Staudacher (BHL/BES) phase, and without it. Even
if the kernel without the dressing phase is by now only of limited interest, we can still ask the
question whether the su(1|1) sector arises from an effective model, in the same way the su(2)
sector appears from the reduction of the Hubbard model at half-filling. The large g limit of the
su(1|1) sector should give a hint for the underlying degrees of freedom. In the su(2) sector, which
was solved in [37,38], the degrees of freedom are magnons, which become free as g . Our
result in the su(1|1) sector shows that these degrees of freedom do not correspond to a free system, and they are rather difficult to characterize. Two thirds of the excitations are of the type
giant magnon, with finite momenta, and one third are of the type plane wave, with momenta
of order 1/g, see also the numerical results of [39]. We find for the energy without the dressing
factor
8 ln 2
4
gL,
Esu(2) = gL.
(1.8)
The su(2) energy is the g limit of the exact result [37,38]. When taking into account the
BHL/BES phase, almost all roots of the Bethe equations, both for the su(2) and su(1|1) sector,
fall into the region near-flat space regime, with momenta of order 1/ g. This phenomenon
was first discussed in [11], where the 1/4 behavior was obtained from the BA equations. The
integral equations are able to reproduce the leading result for the anomalous dimensions of the
corresponding states, which were also obtained numerically and analytically in [39,40]

d
d
= 2gL,
Esu(2)
= gL.
Esu(1|1)
(1.9)
Esu(1|1) =
We also analyzed the strong coupling limit of the equation derived by Eden and Staudacher
(ES) [29] for the anomalous dimension of the twist-two operator without and with the dressing
phase (BES equation). Similarly to the approaches [23,24], we obtain that the strong coupling
limit of the ES equation is pathological. The BES equation has a better behavior, and we are
able to reproduce the leading term of the density obtained recently by Alday et al. [25] using the
Fourier space representation,
1
for |u| < 2g,
4g 2
1
(1 cosh s/2) for u = 2g coth s, |u| > 2g.
> (u) =
4g 2
< (u) =
(1.10)
This solution reproduces the leading behavior of the universal scaling function (1.2).
For the next order in the density, an important fraction of the roots lie near the points u 2g.
We show that without properly considering the scattering of the magnons in this region, the
density is non-normalizable. We leave the fine analysis of the near-flat space region for future
work.
The paper is organized as follows: in Section 2 we write down the integral equations corresponding to the three sectors, in Section 3 we present in detail the elliptic parametrization and
417
we take the strong coupling limit of one of the building blocks, the su(1|1) kernel. In Section 4
we compute the energy of the highest states for the su(2) and su(1|1) without the dressing kernel
and discuss the leading order solution of the EdenStaudacher equation. In Section 5 we give an
integral representation of the dressing kernel and we check that it reproduces the already known
results at strong coupling. We also give the strong coupling limit of the AFS phase [11]. The
integral equations with the dressing phase equations are considered in Section 6.
2. Bethe ansatz equations in integral form
If we consider highly excited states, with a large number of magnons, the Bethe ansatz equations can be solved by transforming them into integral equations. These integral equations are
particularly simple when the number of magnons is maximal, such that there are no holes in
the magnon distribution.4 Further simplification arise if we restrict ourselves to one of the three
rank-one sectors, su(1|1), su(2) or sl(2). The su(1|1) comprises one complex boson Z and one
fermion U , and the state with the maximum number of fermions
Tr U L
is, at least at weak coupling, the one with highest energy.
In the su(2) sector, corresponding to two complex bosons Z and X, the state with highest
energy is, again at small coupling g, the antiferromagnetic state. This is, in the case of even
length, the singlet state
Tr(ZX)L/2 + .
The sl(2) sector corresponds to combinations of one boson Z and the covariant derivative D.
The interesting operators are the so-called twist-L operators with spin S,
Tr D S Z L .
The equations in the three rank-one sectors can be written in the compact form [8,30]

+ L
M
1 g 2 /xk+ xl xk+ xl i(uk ,ul )
xk
=
e
,
xk
1 g 2 /xl+ xk xk xl+
l
(2.1)
where = 1, 0, 1 for su(2), su(1|1) and sl(2), respectively. The variables x are defined by

1
x = x(u i/2).
x(u) = u 1 + 1 4g 2 /u2 ,
2
It is often convenient to work with the momentum p, the physical variable of the magnon, defined
by

p
p
1
1 + 16g 2 sin2 , or x + /x = eip .
u(p) = cot
(2.2)
2
2
2
We are going to consider both the case where the phase (uk , ul ) is zero and the case where it
equals the BHL/BES dressing phase [13,14].
4 The integral equation for the highest state in the su(1|1) sector was written down by one of us and M. Staudacher,
[41] and in [44], the equation for the antiferromagnetic state in the su(2) sector was solved in [37], and the equation in
sl(2) sector, or the EdenStaudacher equation, was derived in [29].
418
Generically, the Bethe equations for the rank-one sectors can be written as
eipk L =
ei(uk ,ul ) ,
(2.3)
l=1
where (uk , ul ) is the scattering phase for two magnons. Let us exemplify the derivation of the
integral equation on the su(1|1) case, which is the simplest one. We consider for convenience
L odd. The total number of magnons is M = L. When g = 0, Eqs. (2.1) correspond to a free
fermion system, with the occupation numbers
L1
L1
,...,
.
(2.4)
2
2
When g increases, the momenta start to evolve, according to the Bethe ansatz equations (2.1)
(0)
pk = 2k/L,
k=
k
1
pk
(uk , ul ).
= +
2
L 2L
M
(2.5)
l=1
When the length of the chain is very large, we can take the continuum limit. We introduce the
variable t = k/L and the density of rapidities (u) = dt/du. The derivative with respect to u
of Eq. (2.5) gives
1 dp
+
(u) =
2 du
du K(u, u )(u ),
(2.6)
where the kernel is the derivative of the scattering phase

1 d
(2.7)
(u, u ).
2 du
The derivation of the su(2) equation is similar, with the difference that the maximum number of
magnons is L/2 instead of L. This derivation follows closely the analysis of the antiferromagnetic state in the XXX model [42]. A more sophisticated treatment which is able to take into
account the finite size corrections, was used in [43] to study the antiferromagnetic state in the
BDS model [30] and in the Hubbard model.
The Bethe integral equations (2.1) can be formulated in terms of three basic kernels. The first
one is the kernel K(u, u ), corresponding to the su(1|1) case without the dressing phase. The second one is the su(2) kernel Ksu(2) (u, u ), and the third one is the dressing kernel, corresponding
to the dressing phase (u, u )

1 d
g2
g2

K(u, u ) =
(2.8)
ln 1 +
ln 1
,
2i du
x (u)x (u )
x (u)x + (u )

1 d
1
u u + i
1
,
ln
=
Ksu(2) (u, u ) =
(2.9)

2i du
uu i
(u u )2 + 1
1 d
Kd (u, u ) =
(2.10)
(u, u ).
2 du
When the dressing phase is taken into account, the complete kernel is
K(u, u ) =
Ktot (u, u ) = (1 )K(u, u ) + Ksu(2) (u, u ) + Kd (u, u )

K(u, u ) + Ksu(2) (u, u ).
(2.11)
419
It will be sometimes convenient to work with the phase associated with the kernel K
1
(2.12)
u (u, u ), = (1 ) + .
2
The sl(2) sector is special in the sense that the number of magnons, also called spin, M = S,
is not bounded, even if the length is finite (L = 2 for the twist-two operator). The lowest state
with S magnons can be characterized by the numbers
K(u, u ) =
nk = k + sign(k),
k = (S 1)/2, . . . , (S 1)/2.
(2.13)
The corresponding integral equation is given, again in the absence of the dressing phase, by
2
1
(u) = (u) +
S
(u )
+2
du
(u u )2 + 1
du K(u, u )(u ),
(2.14)
where the term proportional to (u) comes from the distribution of the mode numbers nk . The
limit g = 0 of this equation is singular, and it was solved in [31]. The solution for large S is

1
1 + 1 4u2 /S 2 2 ln S

0 (u) =
(2.15)
=
ln
+ O(1/S).
S 1 1 4u2 /S 2
S
Eden and Staudacher [29] chose to separate the density (u) into the g = 0 part, 0 (u), and a
fluctuation (u)

1
2 4 ln S
2 4 ln S
(u) + O(1/S) = 2g
(u)
(u) = 0 (u) 2g
+ O(1/S).
S
S
4g 2
(2.16)
From (2.14), the equation satisfied by the density fluctuation (u) is
1
(u) =
(u )
+2
du
(u u )2 + 1

du K(u, u ) (u )

1
.
4g 2
(2.17)
This is essentially the equation derived by Eden and Staudacher [29], with the inhomogeneous
term written differently. This way of writing has the advantage to show that the separation of
(u) into 0 (u) and (u) makes sense, at large g, if K(u, u ) is sufficiently well behaved at
large u . As we shall see, if K(u, u ) = O(1) for large g and |u | > 2g, Eq. (2.17) will have only
the trivial solution,5 since at leading order in 1/g one can write

du K(u, u ) (u )
1
4g 2

= 0.
(2.18)
In this case, the separation of the density into a free part and a perturbation is inconsistent. The
EdenStaudacher kernel is exactly of this type, and this might be the reason the strong coupling
limit of the EdenStaudacher equation is so pathological [23,24]. The BES kernel, on the other
hand, vanishes at the order O(1) for |u | > 2g, and the strong coupling limit of the solution of
the BES equation is well behaved.
5 At least if the kernel is non-degenerate.
420
Fig. 1. The contour C in the x plane for three real values of close to 0.
3. Elliptic parametrization of the kernels

3.1. The elliptic map
In the strong coupling regime, it is convenient to rescale the variables u and x in order to
eliminate the overall factors of g. We will use the rescaled variables throughout the rest of the
paper
u=
uold
,
2g
x =
xold
,
g
=
1
.
4g
(3.1)
The function
x(u) = u +

u2 1
(3.2)
has a branch cut along the interval [1, 1]. Since the Bethe equations involve the shifted variables
u i, they contain two symmetric cuts, [1 i, 1 i] and [1 + i, 1 + i].
The cuts can be removed by introducing a global parametrization. We eliminate the parameter
u from
u u i =

1
x + 1/x
2
to obtain the following relation between x + and x

+
1
x x 1 + = 4i.
x x
(3.3)
(3.4)
If we require that the momentum p is real, then x + and x are complex conjugate and u real.
Then the condition (3.4) for x = x + and x = x define a contour C in the complex x-plane. The
contour consists, for g real, of two connected components, C and C0 , which are exchanged by
the particleantiparticle transformation x 1/x. We denote by C the component that contains
the point x = ; the other component contains the point x = 0. In the limit 0 the contours
C and C0 develop cusps where they touch (Fig. 1). At this point the contour C is the union
of the real line and the unit circle. A second singular value is = i. Here the two connected
component join into one (Fig. 2).
421
Fig. 2. The contour C for three imaginary values of close to i.
The equation for the contour C can be written conveniently in terms of the momentum p and
a complementary parameter
x = e(ip)/2 .
(3.5)
In these variables the relation (3.4) reads

p
sinh =
2
2
and the parameter u is expressed as
sin
(3.6)
(3.7)
cosh .
2
2
In the following, we consider only positive values of , and p [0, 2]. This insures that u in
(3.7) takes all the real values once and only once.
The relation (3.6) is resolved by elliptic map6 with modulus k defined as
u = cos
=
k
k
or
k=
1
1 + 2
(3.8)
Then and p are parametrized by the Jacobi elliptic amplitude function

p(s) = 2 am(K s, k),
(s) = i 2i am(iK s, k).
(3.9)
When the elliptic parameter s sweeps the interval [0, 2K], the momentum p increases from 0
to 2 . The symmetries of p and are
p(s) = p(s),
(s) = (s) + 2i,
p(s + 2K) = p(s) + 2,
(3.10)
(2K s) = (s).
(3.11)
The functions we will work with are actually expressed in terms of Jacobi elliptic functions
1 dn s
,
eip/2 = cd s ik sd s,
k sn s
s = 2 cs s.
s p = 2k nd s,
e/2 =
(3.12)
6 Our map is related to that of [10] by a GaussLanden transformation k

Janik = 2 k /(1 + k ).
422
Fig. 3. Left: The function u(s) for = 103 on the interval [0, 2K]; Right: The functions p(s) (in blue) and (s) (in red)
for = 103 .
The contours C and C0 are parametrized respectively by s I and s I0 , where

I = [0, 2K],
I0 = [2iK , 2K + 2iK ].
(3.13)
The original rapidity variable u is parametrized as

u(s) =
1 cn s
.
k sn s dn s
(3.14)
The map (3.14) obeys the symmetries u(s) = u(s) = u(s + 2K) = u(s + 2iK ), as well as
u(s)u(K s) = 1 + 2 .
(3.15)
The real axis in the u-plane is the image of the interval I . The functions x (s) are periodic in
s with periods 2K and 4iK and have the symmetries
x + (s) = x + (K iK s) = x (s),
x (s + 2iK ) =
1
x (s)
(3.16)
The first symmetry preserves the contours C and C0 , while the second symmetry exchanges
them. Note that the factors e/2 and eip/2 are themselves antiperiodic in s s + 2K. This is
a little complication, due to the fact that we are working with trigonometric functions of p/2,
which are periodic with period 4 . However, in the physical variables, they always appear in
combinations which are periodic with period 2 . As we already mentioned, we prefer to work
with p in the interval (0, 2).
In the limit 0, where K and K /2, the elliptic parametrization degenerates7
and the variables x and u become hyperbolic functions of s. As the plot of u(s) in Fig. 3 suggests, the parametrization interval splits into two different pieces, corresponding to two different
regimes. By periodicity,8 we can choose the parametrization interval as [ 12 K, 32 K]; then the
first regime corresponds to the interval [ 12 K, 12 K], which becomes the whole real axis when
K . The second regime, corresponding to the interval [ 12 K, 32 K] can be mapped also to the
interval [ 12 K, 12 K] by the transformation
s K s.
(3.17)
7 Another point where the torus degenerates into a cylinder is = i, or g 2 = 1/16, cf. Fig. 2.
8 As already explained, the variables u and x are periodic with period 2K, and we can safely do the shift s s 2K
for them, but for the variables p, we have to remember that the point 12 K was originally 32 K.
423
The first interval corresponds to values |u| > 1, while the second corresponds to |u| < 1. We
therefore introduce the suggestive notation

1 1
1 3
I < = K, K
I < = K, K ,
(3.18)
2 2
2 2
and use the asymptotic formulas valid in the limit 0 and for s I >
sn s = tanh s,
sn(K s) = 1,
cn s = 1/ cosh s,
cn(K s) = k sinh s,
dn s = 1/ cosh s,
dn(K s) = k cosh s.
(3.19)
The next terms in the expansion in are given in Appendix A.

In the hyperbolic limit, the expression (3.14) for u becomes, in the two different regimes,
u> (s) = u(s) coth s,
u< (s) = u(K s) tanh s.
(3.20)
The points s = K/2,

which are the fixed points of the transformation s K s, correspond
to the points u = 1 + 2 . In the strong coupling limit, 0, these are the points u = 1,
and they correspond to meeting points s of the two parametrizations (3.20). These points
will be very important, and we need a better parametrization for them. In the strong coupling
limit the vicinity of these points is parametrized by (Appendix A)

1
u K + s = 1 sinh 2s + O 2 .
(3.21)
2
The expressions for the momentum p and its counterpart in these regimes are
sin p/2 = es ,

sinh /2 = es ,
(3.22)
where the upper sign correspond to momenta close to 0 and the lower sign to momenta close
to 2 . A similar parametrization was used in [28].
3.2. Strong coupling limit of the su(1|1) kernel
We saw that in the strong coupling limit the integration interval splits naturally into two
domains, |u| < 1 and |u| > 1, which are the images of the intervals I < and I > in the s parametrization. As it is clear from Fig. 1, in the first domain the complex variable x = x + becomes
asymptotically unimodular, while in the second domain it becomes asymptotically real. For
0 the relations (3.6) and (3.7) give
0,
u = cos p/2 if |u| < 1,
(3.23)
p 0,
u = cosh /2 if |u| > 1.
(3.24)
Passing to the elliptic parameter s we write the integral equation (2.6) as

1
p (s) + ds1 K(s, s1 )(s1 ),
(s) =
2
(3.25)
where

(s) = u (s)(u),

K(s, s1 ) = u (s)K(u, u1 ).
(3.26)
424
The first term in the r.h.s. of (3.25) changes sign because the derivative u (s) is negative. The
kernel is equal to the derivative

1
1
1
1
1 d
ln 1 e 2 (+1 ) ei 2 (pp1 ) ln 1 e 2 (+1 ) ei 2 (pp1 )
2i ds
1 p (s) sinh 12 ( + 1 ) (s) sin 12 (p p1 )
1
p (s)
.
=
(3.27)
4
4
cosh 12 ( + 1 ) cos 12 (p p1 )
K(s, s1 ) =
Let us mention here that the piece p (s)/4 has a physical meaning: it corresponds to a change
of the effective length of the chain. Every magnon increases the effective length of the chain
by 1/2. This is particularly clear on the discrete equations, where, at strong coupling, the Bethe
equations become, as it was noticed in [44]
eipk (L+M/2) = 1,
(3.28)
under the condition that all pk are finite.

It is now easy to take the strong coupling limit using the asymptotic expressions for p(s) and
(s), which can be found in Appendix A. The asymptotic form of the kernel is given by four
different analytic expressions depending on whether its arguments are in the interval I > or I < .
After applying the redefinition (3.17) to the arguments that belong to the second interval, we can
write the result as a 2 2 matrix kernel
>>

K (s, s1 ) K >< (s, s1 )
K(s, s1 )
K(s, K s1 )
(3.29)
:=
.
K <> (s, s1 ) K << (s, s1 )
K(K s, s1 ) K(K s, K s1 )
The arguments s, s1 are defined in the interval I > , which extends to [, ] when 0.
Similarly the density splits into two components,

>
(s)
(s)
=
(3.30)
(K s)
< (s)
satisfying, in the limit 0, the normalization condition9

ds < +
ds > = 1.
(3.31)
Eq. (3.25) now takes the form

(s) =
>

ds1 K
>>
(s, s1 ) (s1 ) +
>
ds1 K >< (s, s1 ) < (s1 ),
p (s)
(s) =
+
2
<

ds1 K
<>
(s, s1 ) (s1 ) +
>
ds1 K << (s, s1 ) < (s1 ).
The matrix elements of the kernel (3.29), evaluated in Appendix A, are

K << (s, s ) =
(3.32)
1
1
(s s ),
2 cosh s
9 This is true only for the kernel without the dressing phase, at the leading order.
(3.33)
425
1
1
1
1
,
2 cosh s 2 cosh(s s )
1
1
,
K >< (s, s ) =
2 cosh(s s )
K <> (s, s ) =
K >> (s, s ) = 0.
(3.34)
Note that in the hyperbolic limit all four integration kernels depend on the difference of the
arguments. This is a considerable simplification, since the equation can be now solved using
Fourier transform.
In the third regime, which corresponds to the points around u = 1 the momenta are parametrized by (3.22). This regime will be very important for evaluation of the density with the
BHL/BES kernel. The appropriate parametrization is given by (3.21) and (3.22). The only component of the undressed kernel that survives at the leading order is when either u, u 1 or
u, u 1
K ++ (s, s ) = K (s, s )

sinh(s s ) + 1
1
+O .
=

4 cosh(s s ) cosh(s + s ) sinh(s + s )
(3.35)
If only one of the arguments is in the intermediate region, the kernel is of order .
One might argue that there are many intermediate regimes, where p scales like an arbitrary
power of ,
,
p 1 ,
(0, 1).
(3.36)
However, if = 1/2, one of the variables p and dominates the other, and these regimes are
properly taken into account in the regions . Only for = 1/2 the quantities p and are of the
same order of magnitude and this is why we have to consider this case separately.
4. Solving the integral equations in the strong coupling limit without the dressing kernel
4.1. The su(2) case
The su(2) kernel is the only one which is of difference form, therefore it can be exactly solved
for any value of g [37,38]. The energy of the antiferromagnetic state is identical to the energy of
the ground state for the Hubbard model at half filling [46]. Here, we would like to comment on
the strong coupling limit of this solution. The integral equation is particularly simple, since the
kernel becomes simply the delta function
Ksu(2) (u, u ) = lim
0
1
2
= (u u ).
(u u )2 + 4 2
(4.1)
The integral equation reduces then to

2(u) =
1 dp
,
2 du
or (p) =
1
.
4
(4.2)
The solution corresponds to L/2 roots distributed uniformly between p = 0 and 2 . In this case,
the density is normalized to 1/2, because the total number of magnons is L/2. All roots of the
426
Bethe equations are inside the interval |u| < 1 and the energy is
2
Esu(2) = 4gL
dp | sin p/2|(p) =
4gL
.
(4.3)
Physically, the excitations are the (usual) magnons. The solution obtained above shows that,
in the strong coupling limit, the magnons are free, except for the statistical repulsion (4.1). The
situation is very much similar to that of the HaldaneShastry model [47,48], where the magnons
are also free up to the statistical repulsion, translated into the rule that two magnons cannot
occupy successive momenta. Let us remind that the scattering phase for particles with purely
statistical interaction is [49]
(p, p ) = ( 1) sign(p p ),
(4.4)
which is consistent with (4.1), the statistical parameter of the magnons being = 2. The only
difference with the HaldaneShastry spin chain is that here the magnons have the dispersion
relation
E(p) = 4g| sin p/2|,
(4.5)
which is typical for a finite-difference Hamiltonian, while the HaldaneShastry magnons have
the dispersion relation
EHS (p) = p(2 p).
(4.6)
A natural candidate for a model with purely statistical interaction and with trigonometric dispersion relation is the RuijsenaarsSchneider model [50]. This suggests the existence of a spin
model which would describe the spin sector of the half-filled Hubbard model for any value of g
and which might be a multi-spin deformation of the Inozemtsev model [51,52].
4.2. The su(1|1) case
The solution of the su(1|1) sector is considerably more involved, but its leading order still can
be obtained in closed form. Since now the roots of the Bethe equations occur both in the regions
|u| < 1 and |u| > 1, the integral equation splits up into two coupled equations for the densities in
the two regions
3
1
1
2 (s) =
2 cosh s 2
<
1
(s) =
2
>
ds
ds
< (s )
.
cosh(s s )
> (s )
,
cosh(s s )
(4.7)
(4.8)
Substituting the second equation into the first we obtain a new equation where the kernel is of
difference form

s s
3
1
1
<

ds
< (s ),
(s) =
(4.9)
4 cosh s 4 2
sinh(s s )
427
and therefore is solved by Fourier transformation. In Fourier space the equation becomes algebraic,10
< (t) =
2
3
< (t),
2
4 cosh(t/2) 4 2 cosh2 (t/2)
(4.10)
with the solution

6 cosh(t/2)
.
1 + 8 cosh2 (t/2)
Transforming back to the s-variable, we get
< (t) =
1 cos 2as/
,
< (s) =
2 cosh s
a = ln 2 0.34657.
For the density of roots (u) we obtain in the interval |u| < 1

< (s)
a 1+u
1
1
<
(u) =
cos
=
ln
.
|u (s)|
1u
2 1 u2
(4.11)
(4.12)
(4.13)
(4.14)
To evaluate the function > (u), we make the change of variable u = coth s, which is appropriate for |u| > 1, so that we get
1
(s) =
2 2 2
>
ds cos(2as /)
1 sin(2as/)
=
,

cosh s cosh(s s )
sinh s
(4.15)
or, in the initial variables,

a u+1
1
1
sin
ln
.
(4.16)
u2 1
u1
The result shows a singularity at the points u = 1; moreover, it can take negative values. This
happens at values of 1 u 105 and is presumably due to the roots around u = 1 which
were not taken properly into account. Except for these oscillations, the result reproduces nicely
the solution obtained by numerical integration of the Fredholm
equation.11

The total normalization of the density is insured, since < du < (u) = 2/3 and > du > (u) =
1/3. So two third of the roots of the Bethe equations fall in the region |u| < 1 and one third in
the intervals |u| > 1. This phenomenon is also present in the numerical solution of [39]. Finally,
the energy of the state is given by
> (u) =
1
Esu(1|1) = 4gL

du < (u) 1 u2
(4.17)

2 2gL
cos 2as/
16a
=
ds
gL.
=
2
cosh s
(4.18)
10 In order to keep the notation simple, we are going to use the same symbol for the Fourier transforms < (t), > (t)
of the densities.
11 We thank M. Staudacher for supplying us the solution obtained by numerical integration.
428
Fig. 4. Density profile predicted by the analytic solution (4.14), (4.16). Only the u > 0 part is shown. The oscillations do
not show yet at the chosen scale.
This result is consistent with that of Beccaria and Del Debbio [39]. We obtain for their constant
cL the value cL = 2 ln(2)/ 2 0.14046.
4.3. The sl(2) case
The dimension of the twist two operator was originally supposed to be determined by the
solution of EdenStaudacher equation [29]
1
(u) =
du
2
(u ) + 2
(u u )2 + 4 2

1
du K(u, u ) (u )
.
2g
(4.19)
The anomalous dimension is simply given by the normalization of the density fluctuation
(u) [53]

f (g) = 16g
(4.20)
du (u),
or alternatively by the formula [29]

f (g) = 8g
1 2g

du (u)
i
i
x + (u) x (u)

.
As explained in Section 2, in the limit = 1/4g 0 Eq. (4.19) becomes

1

= 0.
du K(u, u ) (u )
2g
(4.21)
(4.22)
429
Obviously, this equation has as solution12

(u) =
1
2g
(4.23)
on the whole real axis. This solution, first obtained in [29], seems to be correct in the interval
|u| < 1, since it insures the vanishing of the O(g 2 ) term in f (g)

f (g) = 8g
1
1 4g

2
du (u) 1 u + O(1/g) = O(g).
<
(4.24)
In the interval |u| > 1, although it has the right scaling in g, the solution (4.23) is not integrable
and it gives formally an infinite value for the anomalous dimension of the twist-two operator. As
mentioned in Section 2, the density (4.23) exactly compensate the leading order in S in 0 (u)
everywhere in u
4g ln S
+ .
(4.25)
S
Unfortunately, as we show below, (4.23) is the unique solution to the ES equation at strong
coupling. We can write the equations as

< (s )
1
1
1
ds
=
,

2
2
cosh(s s ) 8g cosh (s/2)

> (s )
1
1
A
1
1
ds
+ < (s) =
,
+
(4.26)

2
2
cosh(s s )
4g cosh (s) 2 cosh(s)

where A = ds ( < (s) + > (s)) is the normalization of the density. In Fourier transformed
form, the equations become
0 (u) =
t
< (t)
=
,
2 cosh t/2 2g sinh t
t
A
> (t)
+ < (t) =
+
.
2 cosh t/2
4g sinh t/2 2 cosh t/2
(4.27)
The first line is readily solved as

< (t) =
1
,
2g sinh t/2
or < (s) =
while the solution for the second is

t
,
> (t) = A
2g tanh t/2
or
> (u) = A(u ) +
1
1
,
2g cosh2 s
or > (s) = A(s) +
or < (u) =
1
,
2g
(4.28)
1
1
,
2g sinh2 s
1
.
2g
(4.29)
The normalization A is not fixed by the equation.

All other approaches tried to compute the strong coupling limit from the EdenStaudacher ran
into some pathology as well: the result obtained by Kotikov and Lipatov [23] oscillates strongly,
12 Here we use the variable u rescaled by 2g, whence the extra factors of 2g in the density.
430
the result obtained numerically in [24] may not converge to a straight line in g as expected13
and the numerical solution for (u) in [29] does not seem to converge to a value with definite
normalization when g . We believe that all these pathologies are related to the separation
of the density (u) into 0 (u) and (u). Rather surprisingly, this pathology disappear when the
dressing kernel is taken into account, as we will show in Section 6.
5. The dressing kernel
5.1. Integral representation
Beisert, Eden and Staudacher proposed an expression for the dressing factor in the Bethe
ansatz equations. This dressing factor translates into a correction to the kernel in the integral
equation. Their formula is given in Fourier transformed form; for our purposes an expression in
terms of variables u is more suited. Let us start with the Fourier transform of the su(1|1) kernel14
t ) = (1 sign tt )|t|e(|t|+|t |)
K(t,
Jn (|t|)Jn (|t |)
n
,
|tt |
n>0
which is related to the kernel of [14] by

t ) = (1 sign tt )|t|e(|t|+|t |) K m |t|, |t |
K(t,
2
and can be broken into a symmetric and antisymmetric part
(5.1)
t ) = K + (t, t ) + K (t, t )
K(t,

= (1 sign tt )|t|e(|t|+|t |) K0 |t|, |t | + K1 |t|, |t | .
2
In the ES and BES equations, the kernel is written in the Fourier variables defined only for
positive values of t and t . This is made possible since the densities (u) and (u) are symmetric
under u u, and therefore under t t, for the states under consideration. For generic states,
this symmetry is not present, and this is why it is desirable to have a non symmetric kernel at
hand. The extension of Kd (t, t ) to negative values of t and t is ambiguous, and undoing the
Fourier transform in the magic formula of [14] will be ambiguous as well.
The authors of [25] gave a prescription to define Kd (u, u ) by symmetrizing all the kernels
on the second variable K(u, u ) 12 (K(u, u ) + K(u, u )). We try to avoid symmetrizing the
kernels from the beginning, and give here a different prescription, using instead that the full BES
kernel is antisymmetric with respect to u u on each variable separately. We remind that
the transformations u u and u u translate in the following way on the variable p
u u p 2 p,
(5.2)
u u p p,
(5.3)
while the variable stays unchanged. Of course, these two transformations act differently on the
Fourier transform.
13 We thank the authors of [24] for checking the behavior of their ES solution.
14 Our definition of the kernel differs by a minus sign from the one of [29].
431
It will be simpler to work with the non-symmetrized kernels and impose the antisymmetry
under (5.3) at the end of the computation. We define, for t, t > 0,

K d (t, t ) 4te(t+t ) K c (t, t )

(t+t )
dt K 1 (t, t )
= 4te
4
=
For any values of
t, t ,
4
K d (t, t ) =
t
K 0 (t , t )

e2t 1
e2|t |
K+ (t , t ).
dt K (t, t ) 2|t |
e
1
we can define the dressing kernel as a convolution

e2|t |
dt K (t, t ) 2|t |
K+ (t , t ).
e
1
(5.4)
The dressing kernel K d (t, t ), defined in this way, is proportional to 1 sign tt . Note, however,
that after antisymmetrization under (5.3) this property will not hold anymore. Going back to the
u variables we obtain

Kd (u, u ) = 8
dv K (u, v)K+ (v, u )
4

dv dv K (u, v)h(v v )+ (v , u ),
(5.5)
where
h(u) =
2
2

dt
2
|t|eitu
=
e2|t| 1 2
(/2)2
1
u2 sinh2 (u/2)

(5.6)
and + (u, u ) is the Fourier transform of K + (t, t )/|t|. The function h(u) becomes (u) as
approaches 0. Since the kernels involved in the convolution in (5.5) are of order 0 , the dressing
kernel will be of order 1 . This formula is very close to the one in [25], except that we traded
the difficulty in evaluating the function h(u) for the difficulty in evaluating the phase + (u, v).
Let us now write down the expressions of K+ (u, u ) and K (u, u ), which contain the odd,
respectively even powers in the expansion of the logarithm
1
(1 X+ )(1 + X+ )
u ln
,
4i
(1 X+ )(1 + X+ )
1
(1 X+ )(1 + X+ )
K+ (u, v) =
u ln
,
4i
(1 + X+ )(1 X+ )
1
(1 X+ )(1 X+ )
ln
,
+ (u, v) =
4 (1 + X+ )(1 + X+ )
K (u, v) =
(5.7)
(5.8)
with
X+ =
1
x + (u)x (v)
X+ =
1
x (u)x + (v)
(5.9)
432
Fig. 5. The sequence of contours in the definition of the dressing kernel drown in the x-plane (left) and in the
(p + i)-plane (right).
A strategy to compute the expansion of Kd in the strong coupling limit is to compute the limit of
the auxiliary kernels (5.8), and then perform the integrals, as it was done in [25]. This procedure
becomes rapidly quite involved, since different pieces of the kernel contribute for different orders
in .
In the following, we give yet a different representation for the dressing kernel, which may be
more useful for the strong coupling expansion. Eq. (5.4) can be written as
4
K d (t, t ) =

dt K (t, t )e2n|t | K + (t , t ).
(5.10)
n0
This formula can be easily Fourier transformed back such that we get
Kd (u, u ) = 8

1,n
n,1
dv K
(u, v)K+
(v, u ),
(5.11)
n1
with
m,n
K
(u, v) =
1
u
2i

l>0; even
odd
l
l
1 (+m)
,
(u)x (n) (v) x (m) (u)x (+n) (v)
x
l

x (n) (u) = u in + (u in)2 1.
(5.12)
The variables x (n) appeared in [13], where they play an important role in defining the dressing
phase. They live on a torus with modulus defined by
kn
= n
kn
and they are associated with the bound states of magnons [35].
The expression (5.11) for the dressing kernel involves a sum of integrals along an infinite set
, n = 1, 2, 3, . . . , depicted in Fig. 5, defined in the same way as the contour C ,
of contours Cn
but with replaced by n

Cn
(5.13)
= x C (x x)(1
1/x x)
= 4in, xx
>1 .
433
In the strong coupling limit, the expression (5.11) becomes tractable, since we can evaluate
the sum using the EulerMaclaurin formula

n=1
1
f (n) =

B2k (2k1)
1
f (z) dz f (0)
2k1
(0).
f
2
(2k)!
(5.14)
k=1
We thus extend n to the continuous variable

n = z
(5.15)
and define the contour Cz by

p
(5.16)
sinh = z, > 0.
2
2
The integrals over v and the variable z become a double integral over the domain D in the
complex x-plane spanned by the contours Cz . The domain in question is the upper half plane
minus the unit semi-disk. In terms of the complexified momentum + ip = 2 ln x, the domain
D is the semi-infinite cylinder D = {0 p < 2, > 0}.
The integral can be done without specifying a particular parametrization of the contour Cz if
we express the integrand as a wedge product of differential forms. We first write the basic kernel
as a 1-form
Cz :
sin
K(x, x1 ) =
pp
+
dp
1 sinh 2 1 dp sin 2 1 d
.
+
4
4 cosh +1 cos pp1
2
2
(5.17)
Then the even kernel K+ is a one-form in the basis of dp and d,

K+ (x, x2 ) =
pp
pp
+
+
1 sinh 2 2 cos 2 2 dp sin 2 2 cosh 2 2 d
,
2
cosh( + 2 ) cos(p p2 )
(5.18)
while the odd kernel K is a zero-form in this basis,

K (x1 , x) =
1
1 dp1 sinh( + 1 ) d1 sin(p1 p)
dp1 +
.
4
4
cosh( + 1 ) cos(p1 p)
(5.19)
Also, for the differential dz we have from (5.16)

1
1
sinh(/2) cos(p/2) dp + sin(p/2) cosh(/2) d.
2
2
Then the leading term in the strong coupling limit 0 is given by
dz =
8
Kd0 (x1 , x2 ) =

dz
xCz
8
K K+

(5.20)

K dz K+ .
(5.21)
The evaluation of this integral (the calculation is sketched in Appendix B) leads to the following
expression for Kd0

1
Kd0 (x1 , x2 ) =
u x1 , x2+ + x2+ , x1 + x1 , x2+ x2+ , x1 + c.c. ,

4
1
1
log y
.
(x, y) = x +
(5.22)
x
x
434
This expression must be antisymmetrized with respect to x2 x2 (or, equivalently, p2 p2 ,

2 2 ). After antisymmetrization the result coincides15 with the AFS kernel K AFS (u, v), symmetrized with respect to v v.
5.2. The strong coupling limit from the AFS kernel
To compute the leading order(s) of the dressing kernel, we do not really need the integral
representation derived in the previous subsection. It is more straightforward to use the strong
coupling representation of the dressing phase 12 = (u1 , u2 ) worked out in [13,14]. The latter
can be cast in the form

1
+
+
++
12
+ 12
(1 2) ,
12 12
2

rs
= x1r , x2s , r, s = ,
12
12 =
where the function can be expanded in powers of the inverse coupling constant

=
n (2)n .
(5.23)
(5.24)
n0
To compute the leading term in the anomalous dimension, the AFS term [11] is sufficient,
0 (x, y) =
xy 1 xy 1
ln
,
y
xy
while for the next correction we will need the HernndezLpez [12,54] term

y 1/ y
1
y 1 x 1/y
1 (x, y) =
ln
ln
+ Li2
y+1
x y
y x

y + 1/ y
y 1/ y
y + 1/ y
.
Li2
+ Li2
Li2
y x
y+ x
y+ x
(5.25)
(5.26)
In the following, we give the first non-zero order of the dressing AFS phase. The kernel can be
simply obtained by taking the derivative of the phase divided by 2 , with the right sign. We find,
for the sectors < and >,

2
sinh(1 2 )/4
2 >> (u1 , u2 ) + O 2 ,
sinh 1 /2 sinh 2 /2 sinh(1 + 2 )/4
2 sin p2 /2
2 >< (u1 , u2 ) + O(),
>< (u1 , u2 ) =
sinh 1 /2
2 sin p1 /2
2 <> (u1 , u2 ) + O(),
<> (u1 , u2 ) =
sinh 2 /2

1
sin2 (p1 p2 )/4
<<
(u1 , u2 ) = (cos p1 /2 cos p2 /2) ln 2
+ O(1).

sin (p1 + p2 )/4
>> (u1 , u2 ) =
We remind that is the su(1|1) phase.

15 It is useful to notice that we can substitute with to calculate AFS term in (5.23).
0
(5.27)
435
In the sectors + and , where and p (or 2 p) are of order , we introduce the
parameter , which is of order one, as follows
pi = 2 i ,
i = 2 i1 with i > 0 in the region +,
pi = 2 + 2 i ,
i = 2 i1 with i < 0 in the region .
Then we have

( 1 + 21 )2 + (1 2 )2
1 2
1 12 22 + 22 ln 11
(1 , 2 ) =
2
(1 + 21 )2 + (1 + 2 )2

1 + 21 i(1 2 )
4i ln 11
+O ,
1
1 + 2 + i(1 2 )

(1 , 2 ) = 21 2 + O ,
1
+ O(),
> (1 , u2 ) = 2
sinh 2 /2
2
< (1 , u2 ) = 1 sin p2 /2 + O(1),

2
>
+ O(),
(u1 , 2 ) = 2
sinh 1 /2
2
< (u1 , 2 ) = 2 sin p1 /2 + O(1).

(5.28)
In (5.27) and (5.28) we omitted the signs coming from u being in u > 1 and u < 1. A mnemonic
rule to reproduce these signs is to take sign i = sign ui . In the hyperbolic parametrization with
the variable s, which we give below, the signs are automatically taken into account
>> (s1 , s2 ) = 2 sinh s1 sinh s 2 tanh

s1 s 2
2 >> (s1 , s2 ) + O 2 ,
2
2 sinh s1
2 >< (s1 , s2 ) + O(),
cosh s2
2 sinh s2
2 <> (s1 , s2 ) + O(),
<> (s1 , s2 ) =
cosh s1

2 sinh(s1 s2 )
s1 s2
<<
(s1 , s2 ) =
ln tanh
+ O(1).
cosh s cosh s
2
>< (s1 , s2 ) =
(5.29)
In the regimes where one of the variables is in the region + or , we put i = esi

2
1 + e2(s1 +s2 ) tanh2 s1 s
2
(s1 , s2 ) = (sinh 2s1 sinh 2s2 ) ln
1 + e2(s1 +s2 )

s1 s2
+ 4 arctan e(s1 +s2 ) tanh
(5.30)
+O ,
2

(s1 , s2 ) = 2e(s1 s2 ) + O ,
> (s1 , s2 ) = 2 es1 sinh s2 + O(),

2 es1
< (s1 , s2 ) =
+ O(1),
cosh s2
> (s1 , s2 ) = 2 es2 sinh s1 + O(),
436
2 es2
< (s1 , s2 ) =
+ O(1).
cosh s1
(5.31)
It is important to note that, in what concerns , Eq. (5.30) is not the correct answer, since in
this region all terms n in (5.24) contribute to the O(1) order. This was pointed out by Maldacena
and Swanson [28], who derived an integral representation for the scattering phase in this region.
It is interesting to note that and in Eq. (5.30) depend naturally on the rotated coordinates
s = s1 + s2 ,
s = s1 s2 ,
and that for large values of s, positive or negative, we have

s
++ (s1 , s2 ) 2es tanh 1 sinh s .
2
6. Solving the integral equations in the strong coupling limit with the dressing kernel
Once we have obtained the leading term of the dressed kernels, we are able to compute the
leading contribution to the energy of the states we are interested in. In the su(1|1) and su(2)
sectors, we do not need to obtain the functional form for the density explicitly. This is a fortunate
situation, since in the () regime, where most of the roots are concentrated, we do not know
the kernel in a closed form. The computation is very similar to the one performed in [11], where
the 1/4 behavior for the anomalous dimensions was first derived.
6.1. The su(1|1) dressed case
The results of the numerical and analytical computation by Beccaria et al. [39,40] for the
energy of the highest excited state in the su(1|1) sector can be also derived from the integral
equation. As it was shown in [40], the same results can be obtained using the light-cone Bethe
ansatz equations [55], first derived in the su(1|1) sector by Arutyunov and Frolov [56]. In this
sector, the total kernel is
K = K + Kd .
(6.1)
First, we can show that the densities < and > are subdominant, while the main contribution
comes from . The density in the giant magnons region is given by the equation
< =

1 dp
K< .
+ K<< < + K<> > +
2 ds
(6.2)
By symmetry, we have + () = () and since

0
+ () d +
() d = 0,
we conclude that the terms in (6.2) cancel

< =
1 dp
+ K<< < + K<> > + O(1).
2 ds
(6.3)
437
All the terms except K<< < are of the order 1. Since K<< = O(1/) we deduce that < (s) =
O(). Similarly, we can write

1 dp
K> .
+ K>< < + K>> > +
> =
(6.4)
2 ds
Again the leading term in the sum vanishes and all the terms in the r.h.s. are of the order , so
that > (s) is also of order at most . Therefore the roots of the Bethe equations must concentrate
in the regions , or near-flat space region. In this region, we have
1 dp
+ K< < + K> > (s ) + K + K .
2 d
At the leading order, we are left with

() = d K (, ) ( ) + d K (, ) ( )
=
(6.5)
(6.6)
or, by specializing to the sign + and by integrating over ,

1
k () =
2
+
++
d tot
(, ) + ( )
1
2
0
+
d tot
(, ) ( ),
(6.7)

0 d + ().
+
+
is the counting function,
tot
is defined as tot
=
where
+
+
and at leading order is equal to the dressing part . By multiplication with
2 0 d K
() and integration we get
k + ()
k + ()
1
d k () () =
2
+

d
d ++ (, ) + () + ( )
0
d
d + (, ) + () ( ).
(6.8)
0
+

++
Here we used that (0, ) = (0, ) = 0. Let us remind that
= 0 corresponds to u = .
Due to the antisymmetry of the phase ++ , the first integral in the r.h.s. of (6.8) vanishes, while
the l.h.s. is equal to
1/2
1
d k () () = dk + k + = .
8
0
+
(, ) = 2
Finally, we use that

the form

+
d () =
.
8
(6.9)
and the symmetry + () = () to cast Eq. (6.8) in
(6.10)
This is all we need to compute the leading order in the energy

d
Esu(1|1)
= 4gL
/2
sin p/2(p) dp 4 gL
0
d + () =
2gL.
(6.11)
438
It is remarkable that the energy of this state is of the order 1/4 . It would be interesting to know
if there are states in the su(1|1) sector which have greater energy. This would be possible for
states where the fraction of giant magnons is larger than 1/ g. Presumably, the strong repulsion
among the giant magnons will prevent this phenomenon to happen.
6.2. The su(2) dressed case
The computation of the leading order in the energy for su(2) sector goes similarly to that for
su(1|1), with two differences. First, there is a non-trivial contribution from the su(2) kernel to
++
the leading order of tot
u
su(2) (u, u ) = 2
dv Ksu(2) (v, u ) = 2
dv
u
2
(v
u )2
+ 4 2
= + asym (u, u ). (6.12)
Therefore, the first term in the r.h.s. of Eq. (6.8) does not cancel anymore and we obtain
2
2

1
1
(6.13)
d k + () + () =
d + () +
d + () .
2
Second, the maximum of counting function is 1/4 instead of 1/2, because the total number of
magnons is M = L/2, and

0
1/4
1
d () = dk + = .
4
+
From Eq. (6.13) we obtain

d + () =
,
16
(6.14)
which reproduces16 the value of the energy obtained by Beccaria et al. [39,40]

d
Esu(2)
= gL + O g 0 .
(6.15)
6.3. The sl(2) dressed case

For the sl(2) case, the total kernel with the su(2) piece subtracted, as defined in (2.11), is equal
to
1
(6.16)
u (u, u ).
2
The BES equation for this sector takes the simplest form when written for the shifted density
function
2
(u) (u) ,
(6.17)
K(u, u ) = 2K(u, u ) + Kd (u, u )
16 We thank M. Beccaria for an email exchange which helped us to remedy a discrepancy with their results in the first
version of this paper.
439
which, up to a negative factor, equals the total density at the leading order in S, see (2.16). The
BES equation reads
1
(u) =
2 (u )
du
+
(u u )2 + 4 2
du K(u, u ) (u ).
(6.18)
We are interested in expanding Eq. (6.18) in powers of . At small

2
2
(u )

+ du K(u, u ) (u ) = 0.
+O
du
(u u )2
(6.19)
This equation splits into several equations, which couple the different regions in u. For instance,
if |u| < 1 we have
1
1

2
2
< (u )
> (u )

+ du K<< (u, u ) < (u )
+ du
+O
du
(u u )2
(u u )2
1
>
>
du K<> (u, u ) > (u ) +
d K< (u, ) () = 0.
(6.20)
Eq. (6.18) was analyzed in the first two orders in by the authors of [25]. They did not
consider the regions around u = 1, which can contain a fraction of the roots. Eq. (6.18) suggests
that the -expansion of the density starts at order . The structure of the leading terms (5.28) in the
expansion of the dressing kernel, where half-integer powers of appear, suggests that the density
and the anomalous dimension may involve a half-integer powers of as well. However, these
corrections seem to appear in higher orders in the anomalous dimension, and this may explain
why they have not been seen yet.17 In the following, we are going to ignore the corrections
coming from the vicinity of the points u = 1 and solve perturbatively Eq. (6.18), by considering
the decomposition of the density and the kernel in integer powers of
2
= 1 (u) + 2 2 (u) + 3 3 (u) + ,
K(u, u ) = 1 K1 (u, u ) + K0 (u, u ) + K1 (u, u ) + .

(u) = (u)
(6.21)
Order 0
At the leading order in Eq. (6.20) reads, in simplified notations,
K1 1 = 0.
To this order only the << sector contributes
1
<<
du K1
(u, u ) 1< (u ) = 0.
(6.22)
1
17 As far we can see, the first non-zero order for () is at most 2 , and there may be corrections of order 5/2 to
< (u) and of order 3/2 for > (u). The last of these corrections would induce a term of order g 1/2 in the anomalous
dimension. To decide whether they are here or not we have to go to higher orders in the expansion of the kernel.
440
<< is non-degenerate, therefore we have

It was proven in [25] that the kernel K1
1< (u) = 0, or
equivalently 1 = 2/ .
Order 1
The O() term in the expansion of Eq. (6.20) relates 1 and 2
<< <
K1
2 + K0<> 1> = 0.
The kernel in the second term is antisymmetric in u u
1 u2
1
<>

sign u ,
K0 (u, u ) = u
u 2 1
(6.23)
(6.24)
so that it vanishes upon integration against the symmetric function 1> (u). As pointed out in
[25], the subleading density 2< (u) vanishes as well. It is worth noticing that K0<> (u, u ) is the
sum of two terms, one coming from the su(1|1) kernel and the other from the dressing phase,
<> (u, u ).
K0<> (u, u ) = 2K0<> (u, u ) + Kd,0
Order 2
At this order we have an equation relating 1 , 2 and 3

2
1 (u )
+ K1 3 + K0 2 + K1 1 = 0.
du
(u u )2
The interval |u| > 1

Consider first the region |u| > 1, where the equation takes the form

> (u )
2
du 1 2 + du K1>> (u, u ) 1> (u ) = 0,
(u u )
>
(6.25)
>
1
K1>> (u, u ) 2
u 1
with the kernel

given by (5.29). Eq. (6.25) is a total derivative, so we first
integrate it to

2 1
1

du 1> (u )
(6.26)
+
(u,
u
)
= 0.
1
u u 2
>
As the phase is antisymmetric, there is no integration constant.

At this point, we find it more useful to switch to the variable s, defined by u = coth s, and
(6.26) becomes

1
s s
ds 1> (s ) sinh s coth
= 0.
(6.27)
Now we can reformulate the integral equation (6.26) as a Riemann boundary value problem. We
first rewrite (6.27) in terms of the normalizable density 1 = 1 + 2/

s s
4s
1
.
= 2
2
sinh s
(6.28)
441
Then, after the redefinition

= es ,
r() = r(1/) = 1> (s) sinh s,
(6.29)
the l.h.s. of (6.28) takes the form of a Cauchy integral

d
8
ln

r( )
r(
)
+
2
d
=

( > 0).
(6.30)
The first term on the l.h.s. is actually zero due to the antisymmetry of r().
Eq. (6.30) can be formulated in terms of a boundary condition for the resolvent

R() = R(1/) =
d
r( )
,

(6.31)
which has a cut on the positive axis, namely,

R( + i0) + R( i0) =
8 ln
1
( > 0).
The most general solution of (6.32) with the symmetry R() = R(1/) is of the form
4 ln() + Q() 1/Q(1/)

R() =
,
1/
(6.32)
(6.33)
where Q() is a rational function. The latter is determined by the analytical properties of the
resolvent (no poles outside the positive real axis) and the requirement that the density 1 is
normalizable

du 1 (u) =
>

ds 1> (s) =
d
0
2r()
= finite.
2 1
The only solution with these properties corresponds to Q = 1/2:
4 ln() 12 ( 1/)
R() =
,
1/
and its discontinuity along the positive real axis,

2
2
R( + i0) R( i0)
=
,
r() =
2i
1/ 1
reproduces the solution found by Alday et al. [25]:
< (u) = 1< =
1
,
2g
> (u) = 1> (u) =

1
s
1 cosh
.
2g
2
(6.34)
(6.35)
(6.36)
The total integral of the density,

1
4
1
ds < (s) + > (s) =
+
=
,
g
4g
4g
(6.37)
442
reproduces the correct leading behavior for the twist-two anomalous dimension
f (g) = 4g + .
(6.38)
The interval |u| < 1

Let us assume for the moment that the corrections from the vicinity of the points u = 1 can be
neglected. Then, in the region |u| < 1, the equation will reduce to

> (u )
2
<< <
(6.39)
du 1 2 + K1
3 + K1<> 1> = 0.
(u u )
|u |>1
The kernels in the first and the third term combine to

K1<> (u, u ) +
2
1 sinh s cosh s + 2 (s s ) sinh2 s
1
=
u
.
(u u )2
cosh(s s )
(6.40)
The procedure to obtain the previous expression was the following: we expressed the first two
terms 0 and 1 in the dressing phase in terms of the variables p and and expanded them up
to the desired order in . Then we expressed the results in terms of the variables s, s by using
the relations
u = cos p/2 = tanh s,
u = cosh /2 = coth s .
Integrating the product of the kernel (6.40) with 1> = 2 cosh(s/2) gives zero, so we deduce
that
3< (u) = 0.
(6.41)
Order 3
At this order, Eq. (6.20) gives

2 (u )
2
+ K1 4 + K0 3 + K1 2 + K2 1 = 0.
du
(u u )2
The subleading density 2> (u), which we need in order to compute the subleading term in f (g),
is supposed to be determined by the equation with |u| > 1

> (u )
2
(6.42)
du 2 2 + K1>> 2> = K2>> 1> .
(u u )
>
After integrating with respect to u and passing to the s parametrization, this equation takes the
same form as (6.28), but with different inhomogeneous term

s s
1
1
ds 1> (s ) 2>> (s, s ).
=
2
2 sinh s
(6.43)
The inhomogeneous term in the r.h.s. can be computed from the expression
2>> (s, s ) =
4(sinh s sinh s )2 (s s ) cosh(s s ) sinh(s s )

,
sinh2 (s s )
(6.44)
443
which, after being integrated with 1> (s) = 2 cosh(s/2)|u (s)| gives
1
2 sinh s
ds 1> (s )2>> (s, s ) =
2
sinh s sinh(s/2).
(6.45)
After the redefinition (6.29) we again reduce the integral equation to a Riemann boundary value
problem

r( )
( 1)(2 1)
=
.
R( + i0) + R( i0) = 2 d
22
(6.46)
The solution is
R(s) =
s sinh s cosh(s/2)
,
2
r(s) =
s sinh s sinh(s/2)
,
2 2
which gives for the 2 -correction to the density

2> (s) =
s sinh(s/2)
.
2 2
(6.47)
This correction to the density grows exponentially at s and is not normalizable.18 The
significance of the result (6.47) is that the higher order corrections to the density become large in
the vicinity of the points u = 1, which correspond to s = . Therefore, in order to determine
the next correction to the scaling function f (g) we have to take into account the contribution
coming from the points u = 1 (the regimes ). The equations starting with (6.39) should be
corrected accordingly. We leave this problem for future work.
7. Outlook
We presented a method to treat the strong coupling limit of the Bethe ansatz equations, which
can be used to extract the spectrum of anomalous dimensions. We found it useful to express the
spectral parameter appearing in the Bethe ansatz equations by means of a complexified momentum, with real part p and imaginary part
x = e/2ip/2 .
(7.1)
It is worth noticing that these variables already appeared in [37], where the BDS magnons were
interpreted as bound states of more fundamental, fermionic excitations. There, p is the momentum of the bound states, while is related to the size of the bound state.
Furthermore, the functions of p and are naturally expressed in terms of elliptic functions.
Our elliptic parametrization is a version of that in [10]. In the strong coupling limit, the elliptic parametrization degenerates into a hyperbolic one. In order to take into account properly the
different regimes, we used different expansions of the elliptic functions in terms of hyperbolic
functions, depending on the value of the elliptic parameter s: around s = 0, K and K/2. These
18 Strictly speaking, the maximum value of s to which we should integrate is s
max = K/2 ln . If we stop the
integral at smax , we will obtain a contribution to the universal scaling function of the order ln / , which violates the
hypothesis that the expansion is in integer power of . Also, it signals that the limit 0 is not uniform, and this is in
accordance with the fact that in the near-flat space regime all the terms n contribute [28].
444
regimes correspond to the plane-wave [32], the giant magnons [33,34] and the near-flat
space regions, recently characterized in [28], respectively. It is interesting to note that the momentum p and the energy E/4g of the excitations have, in the plane wave and near-flat space
regime, relativistic-like expressions
p = sinh s,
= (cosh s 1) (plane wave),
s
= 2 es (near flat space),
p = 2 e ,
(7.2)
(7.3)
with s playing the role of the rapidity variable. However, only the positive branch of the energy,
corresponding to particles, appear in the sectors we consider, sectors which should be closed at
any order in perturbation theory. It is important to understand how the antiparticle branch will
appear; most probably by another copy of the sector which joins in. An example of how this
should happen is offered by the su(2) principal chiral model [57,58], where the particles and
antiparticles correspond to the two copies of su(2)-symmetric excitations.
The strong coupling limit of the kernels without the dressing factor is relatively simple. The
giant magnons interact through a delta term, typical for the statistical repulsion. In the su(1|1)
sector, there is also a length-changing term. The excitations of the plane-wave type do not interact with each other, at leading order, but they interact with the giant magnons. The excitations
in the near-flat space regime do not contribute to the leading term.
The strong coupling limit of the dressing kernel is more involved. For the giant magnons
and plane-wave regions, we can do it safely with the strong coupling expansion of the dressing
kernel [1113], at least for the first few orders in . For the near-flat space limit, however, as it
was pointed out in [28], all the terms in the strong coupling series contribute to the leading order.
When we consider the kernels with the dressing phase, the repulsion of the giant magnons is
so large that their density vanishes in the leading order. For su(1|1) and su(2) sectors, most of the
magnons are concentrated in the near-flat space regime. Fortunately, although we do not have an
explicit expression of the kernel in this regime, it is possible to obtain the energy for the highest
excited state at leading order.
The most exciting application of our method is to obtain the strong coupling expansion of the
twist-two operator anomalous dimension and to compare with the string computations [21,22].
Several recent analytical and numerical works [2325] were devoted to this task. In particular,
[25] obtained the leading term, using the Fourier transform representation. They also tried to
obtain the result by a method very similar to ours, and found that the first two orders of the
equation are not sufficient to fix the leading order. Here, we show that it is possible to obtained
the leading order, by going an order higher. We also show that, in order to compute the higher
correction, we have to take into account the contribution from the near-flat space regime.
Note added. An integral representation of the dressing phase, seemingly related to (1.5), was
used in [59] to explore the analytic properties of the S-matrix.
Acknowledgements
D.S. and I.K. thank the Albert Einstein Institute, Potsdam, and Kavli Institute for Theoretical
Physics, Santa Barbara, where part of this work was done, for hospitality. D.S. thanks Matthias
Staudacher for collaboration in early stages of this project. We thank Gleb Arutyunov, Matteo
Beccaria, Marcus Benna, Nick Dorey, Burkhard Eden, Sergey Frolov, Igor Klebanov, Lev Lipatov, Matthias Staudacher and Arkady Tseytlin for helpful discussions and Stefan Zieme for
spotting many typos in the first version of the paper. This work has been partially supported by
445
the National Science Foundation under Grant No. PHY99-07949, by the European Union through
ENRAGE network (contract MRTN-CT-2004-005616) and the by ANR programs GIMP (contract ANR-05-BLAN-0029-01) and INT-AdS/CFT (contract ANR36ADSCSTZ).
Appendix A. The elliptic parametrization
The modulus of the elliptic map is
1

k=
,
k =
.
2
1+
1 + 2
The real and imaginary parts of the complexified momentum
p(s) = 2k
s
dv
,
dn v
1 + k
2
(s) = ln
1 k
(A.1)
s
cs v dv,
(A.2)
have the symmetries

p(2K u) = 2 p(u),
p(u) = p(u),
(2K u) = (u),
(u) = (u) 2i.
(A.3)
The phase in the Bethe equations depends on and p through

1
1
1
1
sinh = ds s,
cosh = ns s,
2
k
2
k
1
1
cos p = cd s,
sin p = k sd s.
2
2
1 dn s
/2
e
=
,
eip/2 = cd s ik sd s,
k sn s
s p = 2k nd s,
s = 2 cs s.
(A.4)
(A.5)
Below we give asymptotic expressions of these functions in the limit k 0, where
2
4

K ,
q = eK/K .
K ln ,
(A.6)

2
16
In this limit it is appropriate to use the expansions of the three elliptic functions in terms of
hyperbolic functions of v = s/2K :

2n

q
()n
sinh 2nv ,
tanh v + 4
sn s =
2kK
1 + q 2n
n=1

2n1
1
n q
()
cosh(2n 1)v ,
cn s =
+4
2kK cosh v
1 + q 2n1
n=1

2n1
1
n q
()
cosh(2n 1)v .
dn s =
(A.7)
4
2K cosh v
1 q 2n1
n=1
In the limit 0 and up to o( 4 ) terms the three functions are given by [45]
sn s tanh s + 2
(sinh s cosh s s)
4 cosh2 s
446

1
s
2 tanh s
cn s

sinh s
,
cosh s
4
cosh s

tanh s
s
1
+ 2
sinh s +
.
dn s
cosh s
4
cosh s
(A.8)
We used only the leading term; it gives a good approximation for s I > = [ 12 K, 12 K]. Similarly, for the interval s I < we can use the approximation
sn(K s) cd s 1,
cn(K s) k sd s sinh s,
dn(K s) k nd s cosh s.
(A.9)
For the functions that enter in the definition of the kernel we find:
(a) In the interval I > :
2
1
,
=
,
sinh s
sinh s
sin(p/2) = k sinh s,
s p = 2k cosh s,
cosh(/2) = coth s,
sinh(/2) =
cos(p/2) = 1,
1
u =
,
u = coth s,
sinh2 s

u+1
s
/2
s
e =
.
e = coth ,
2
u1
(b) In the interval I < , after replacing s K s:
cosh(/2) = 1,
sin(/2) = k cosh s,
(A.10)
s = 2k sinh s,
cos(p/2) = tanh s,
sin(p/2) = 1/ cosh s,
s p = 2/ cosh s,
1
u =
,
u = tanh s,
cosh2 s

1u
1 + ies
ip/2
s
=
,
e =
.
e
(A.11)
s
1 ie
1+u
Plugging these expressions in (3.27) we evaluate the kernel in the four possible regimes. In the
sector s, s1 I > the numerator in (3.27) is of order , while the denominator remains finite.
Therefore
K >> (s, s1 ) = 0.
(A.12)
For the non-diagonal elements of (3.29) we obtain

1
sin(p1 /2)
1
1

,
4 cosh(/2) cos(p1 /2) 2 cosh(s s1 )
1
1
p sinh(1 /2)
p
K <> (s, s1 ) = K(K s, s1 )
4
4 cosh(1 /2) cos(p/2)
1
1
1
1

.
2 cosh s 2 cosh(s s1 )
K >< (s, s1 ) = K(s, K s1 )
(A.13)
Finally, if both arguments are in I < , then the kernel K << (s, s1 ) = K(K s, K s1 ) vanishes
except near the double pole of the denominator at s = s1 , where it can be approximated by a
447
delta-function:
p
1
1
p
4
2 2 + sin2 (p p1 )/2
1
p |s p|(p p1 )

4
1
1
(s s1 ).

2 cosh s
K << (s, s1 ) = K(K s, K s1 )
(A.14)
The regime s K/2

Assume that s = K/2 + y, where y
functions for the shifted argument:
sn(s 12 K)
sn 12 K
cn(s 12 K)
cn 12 K
dn(s 12 K)
dn 12 K
. We first evaluate the three basic Jacobi elliptic
k sd cn
(y),
cn2 +k sn2
cn sn dn
(y),
cn2 +k sn2
dn (1 k ) sn cn
(y).
cn2 +k sn2
When k 0, the values of the three functions at s = 12 K are

1
1
sn K = 1/ 1 + k 1 ,
2
2

1
cn K = k /(1 + k ) ,
2

1
dn K = k ,
2
(A.15)
(A.16)
so that

1
1
sn K + y = 1 e2y ,
2
2

1
cn K + y = ey 1 sinh2 y ,
2

1
dn K + y = ey (1 sinh y cosh y).
2
(A.17)
From here one finds for the asymptotics of the functions p(s), (s) and u(s) for s = K/2 + y:

1
sinh(/2) = ey 1 + cosh 2y ,
2

y
1
1 cosh 2y ,
sin(p/2) = e
2
2
u = 1 sinh 2y + o .
(A.18)
448
Appendix B. Evaluation of Kd0

Here we will evaluate the integral (5.21) for the strong coupling limit of Kd . We have
p
K dp1 + K d1
,
4
p
K+ dp + K+ d
,
K+ (x, x2 ) =
2
1
1
dz = Z p dp + Z d,
2
2
where
K (x1 , x) =
K =
(B.1)
cos(p p1 ) e|+1 |
,
H1
pp2
2
sinh +
2 cos 2
,
H2
Z p = sinh(/2) cos(p/2),
K =
sin(p1 p)
,
H1
+2
2
sin pp
2 cosh 2
,
H2
Z = sin(p/2) cosh(/2),
K+ =
K+ =
Ha = cosh( + a ) cos(p pa ),
a = 1, 2.
The integral to evaluate is

Kd0 (x1 , x2 ) =
1
2 2
1
4 2
2

dp
2

dp

p
p
d K dp1 + K d1 Z p K+ Z K+

d A(p, ) dp1 + B(p, ) d1 ,
(B.2)
where

A(, p) = cos(p p1 ) e|+1 | G(, p),
B(, p) = sin(p1 p)G(, p),
G(, p) =
sin(p p22 ) sinh( + 22 ) + sinh 22 sin p22

.
(cosh( + 1 ) cos(p1 p))(cosh( + 2 ) cos(p p2 ))
(B.3)
The strategy of calculation is the following: first we take an integral over p, which is of the
form:
2
I=
dp R(cos p, sin p),
(B.4)
where R is a rational function. By symmetrization p p we reduce the integral to

2
I=
0

dp R cos p, sin2 p
(B.5)
449
and, after substituting cos p = t, it can be evaluated by taking the residues.

The integral over becomes, after the substitution = log x, an integral from 0 to 1 of a
rational expression and can also be performed explicitly. As a final result, we obtain
2
d
2
d

1
1
1 +
1
dp A(p, ) = i +
x1 + log
2
x2
x1
1+

1
1
1 +
1
dp B(p, ) = +
x1 + log
2
x2
x1
1+
1
x1+ x2
1
x1+ x2
1
x1+ x2
1
x1+ x2

c.c. ,
(B.6)

+ c.c. .
(B.7)
From here we deduce the expression for Kd0 , presented in (5.22).

References
[1] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super YangMills, JHEP 0303 (2003) 013, hep-th/0212208.
[2] N. Beisert, M. Staudacher, The N = 4 SYM integrable super spin chain, Nucl. Phys. B 670 (2003) 439, hep-th/
0307042.
[3] N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 conformal super YangMills theory,
Nucl. Phys. B 664 (2003) 131, hep-th/0303060.
[4] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200.
[5] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett.
B 428 (1998) 105, hep-th/9802109.
[6] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[7] M. Staudacher, The factorized S-matrix of CFT/AdS, JHEP 0505 (2005) 054, hep-th/0412188.
[8] N. Beisert, M. Staudacher, Long-range PSU(2, 2|4) Bethe ansaetze for gauge theory and strings, Nucl. Phys. B 727
(2005) 1, hep-th/0504190.
[9] N. Beisert, The analytic Bethe ansatz for a chain with centrally extended su(2|2) symmetry, J. Stat. Mech. 0701
(2007) P017, nlin.SI/0610017.
[10] R.A. Janik, The AdS(5) S 5 superstring worldsheet S-matrix and crossing symmetry, Phys. Rev. D 73 (2006)
086006, hep-th/0603038.
[11] G. Arutyunov, S. Frolov, M. Staudacher, Bethe ansatz for quantum strings, JHEP 0410 (2004) 016, hep-th/0406256.
[12] R. Hernndez, E. Lpez, Quantum corrections to the string Bethe ansatz, JHEP 0607 (2006) 004, hep-th/0603204.
[13] N. Beisert, R. Hernndez, E. Lpez, A crossing-symmetric phase for AdS5 S 5 strings, hep-th/0609044.
[14] N. Beisert, B. Eden, M. Staudacher, Transcendentality and crossing, J. Stat. Mech. 0701 (2007) P021, hep-th/
0610251.
[15] A.V. Kotikov, L.N. Lipatov, A.I. Onishchenko, V.N. Velizhanin, Three-loop universal anomalous dimension of the
Wilson operators in N = 4 SUSY YangMills model, Phys. Lett. B 595 (2004) 521, hep-th/0404092.
[16] S. Moch, J.A.M. Vermaseren, A. Vogt, The three-loop splitting functions in QCD: The non-singlet case, Nucl. Phys.
B 688 (2004) 101, hep-ph/0403192.
[17] C. Anastasiou, Z. Bern, L.J. Dixon, D.A. Kosower, Planar amplitudes in maximally supersymmetric YangMills
theory, Phys. Rev. Lett. 91 (2003) 251602, hep-th/0309040.
[18] Z. Bern, L.J. Dixon, V.A. Smirnov, Iteration of planar amplitudes in maximally supersymmetric YangMills theory
at three loops and beyond, Phys. Rev. D 72 (2005) 085001, hep-th/0505205.
[19] Z. Bern, M. Czakon, L.J. Dixon, D.A. Kosower, V.A. Smirnov, The four-loop planar amplitude and Cusp anomalous
dimension in maximally supersymmetric YangMills theory, hep-th/0610248.
[20] F. Cachazo, M. Spradlin, A. Volovich, Four-loop cusp anomalous dimension from obstructions, hep-th/0612309.
[21] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl. Phys.
B 636 (2002) 99, hep-th/0204051.
[22] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002) 007,
hep-th/0204226.
450
[23] A.V. Kotikov, L.N. Lipatov, On the highest transcendentality in N = 4 SUSY, hep-th/0611204.
[24] M.K. Benna, S. Benvenuti, I.R. Klebanov, A. Scardicchio, A test of the AdS/CFT correspondence using high-spin
operators, hep-th/0611135.
[25] L.F. Alday, G. Arutyunov, M.K. Benna, B. Eden, I.R. Klebanov, On the strong coupling scaling dimension of high
spin operators, hep-th/0702028.
[26] A. Rej, M. Staudacher, S. Zieme, Nesting and dressing, hep-th/0702151.
[27] C. Gomez, R. Hernndez, Quantum deformed magnon kinematics, hep-th/0701200.
[28] J. Maldacena, I. Swanson, Connecting giant magnons to the pp-wave: An interpolating limit of AdS(5) S 5 , hepth/0612079.
[29] B. Eden, M. Staudacher, Integrability and transcendentality, J. Stat. Mech. 0611 (2006) P014, hep-th/0603157.
[30] N. Beisert, V. Dippel, M. Staudacher, A novel long range spin chain and planar N = 4 super YangMills, JHEP 0407
(2004) 075, hep-th/0405001.
[31] G.P. Korchemsky, Quasiclassical QCD pomeron, Nucl. Phys. B 462 (1996) 333, hep-th/9508025.
[32] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super YangMills,
JHEP 0204 (2002) 013, hep-th/0202021.
[33] D.M. Hofman, J.M. Maldacena, Giant magnons, J. Phys. A 39 (2006) 13095, hep-th/0604135.
[34] G. Arutyunov, S. Frolov, M. Zamaklar, Finite-size effects from giant magnons, hep-th/0606126.
[35] N. Dorey, Magnon bound states and the AdS/CFT correspondence, J. Phys. A 39 (2006) 13119, hep-th/0604175.
[36] H.Y. Chen, N. Dorey, K. Okamura, On the scattering of magnon boundstates, JHEP 0611 (2006) 035, hep-th/
0608047.
[37] A. Rej, D. Serban, M. Staudacher, Planar N = 4 gauge theory and the Hubbard model, JHEP 0603 (2006) 018,
hep-th/0512077.
[38] K. Zarembo, Antiferromagnetic operators in N = 4 supersymmetric YangMills theory, Phys. Lett. B 634 (2006)
552, hep-th/0512079.
[39] M. Beccaria, L. Del Debbio, Bethe ansatz solutions for highest states in N = 4 SYM and AdS/CFT duality,
JHEP 0609 (2006) 025, hep-th/0607236.
[40] M. Beccaria, G.F. De Angelis, L. Del Debbio, M. Picariello, Highest states in light-cone AdS(5) S 5 superstring,
hep-th/0701167.
[41] D. Serban, M. Staudacher, 2005, unpublished.
[42] L.D. Faddeev, How algebraic Bethe ansatz works for integrable model, hep-th/9605187.
[43] G. Feverati, D. Fioravanti, P. Grinza, M. Rossi, Hubbards adventures in N = 4 SYM-land? Some non-perturbative
considerations on finite length operators, J. Stat. Mech. 0702 (2007) P001, hep-th/0611186.
[44] G. Arutyunov, A.A. Tseytlin, On highest-energy state in the su(1|1) sector of N = 4 super YangMills theory,
JHEP 0605 (2006) 033, hep-th/0603113.
[45] P. Byrds, M. Friedman, Handbook of Elliptic Integrals for Engineers and Physicists, Springer-Verlag, Berlin, 1954,
p. 25.
[46] E.H. Lieb, F.Y. Wu, Absence of Mott transition in an exact solution of the short-range, one-band model in one
dimension, Phys. Rev. Lett. 20 (1968) 1445;
E.H. Lieb, F.Y. Wu, Absence of Mott transition in an exact solution of the short-range, one-band model in one
dimension, Phys. Rev. Lett. 21 (1968) 192, Erratum.
[47] F.D.M. Haldane, Exact JastrowGutzwiller resonating valence bond ground state of the spin 1/2 antiferromagnetic
Heisenberg chain with 1/R 2 exchange, Phys. Rev. Lett. 60 (1988) 635.
[48] B.S. Shastry, Exact solution of an S = 1/2 Heisenberg antiferromagnetic chain with long-ranged interactions, Phys.
Rev. Lett. 60 (1988) 639.
[49] B. Sutherland, Exact results for a quantum many body problem in one-dimension, Phys. Rev. A 4 (1971) 2019.
[50] S.N.M. Ruijsenaars, H. Schneider, A new class of integrable systems and its relation to solitons, Ann. Phys. 170
(1986) 370.
[51] V.I. Inozemtsev, Integrable Heisenbergvan Vleck chains with variable range exchange, Phys. Part. Nucl. 34 (2003)
166, Fiz. Elem. Chast. Atom. Yadra 34 (2003) 332, hep-th/0201001.
[52] D. Serban, M. Staudacher, Planar N = 4 gauge theory and the Inozemtsev long range spin chain, JHEP 0406 (2004)
001, hep-th/0401057.
[53] L.N. Lipatov, Transcendentality and EdenStaudacher equation, Talk at Workshop on Integrability in Gauge and
String Theory, AEI, Potsdam, Germany, 2428 July 2006, http://int06.aei.mpg.de/presentations/lipatov.pdf.
[54] G. Arutyunov, S. Frolov, On AdS5 S 5 string S-matrix, Phys. Lett. B 639 (2006) 378, hep-th/0604043.
[55] G. Arutyunov, S. Frolov, J. Plefka, M. Zamaklar, The off-shell symmetry algebra of the light-cone AdS5 S 5
superstring, hep-th/0609157.
451
[56] G. Arutyunov, S. Frolov, Uniform light-cone gauge for strings in AdS(5) S 5 : Solving su(1|1) sector, JHEP 0601
(2006) 055, hep-th/0510208.
[57] J.A. Minahan, The SU(2) sector in AdS/CFT, Fortschr. Phys. 53 (2005) 828, hep-th/0503143.
[58] N. Gromov, V. Kazakov, Asymptotic Bethe ansatz from string sigma model on S 3 R, hep-th/0605026.
[59] N. Dorey, D.M. Hofman, J. Maldacena, On the singularities of the magnon S-matrix, hep-th/0703104.
A novel quasi-exactly solvable spin chain

with nearest-neighbors interactions
A. Enciso, F. Finkel, A. Gonzlez-Lpez , M.A. Rodrguez
Departamento de Fsica Terica II, Universidad Complutense, 28040 Madrid, Spain
Received 23 April 2007; accepted 2 July 2007
Abstract
In this paper we study a novel spin chain with nearest-neighbors interactions depending on the sites
coordinates, which in some sense is intermediate between the Heisenberg chain and the spin chains of
HaldaneShastry type. We show that when the number of spins is sufficiently large both the density of
sites and the strength of the interaction between consecutive spins follow the Gaussian law. We develop an
extension of the standard freezing trick argument that enables us to exactly compute a certain number of
eigenvalues and their corresponding eigenfunctions. The eigenvalues thus computed are all integers, and
in fact our numerical studies evidence that these are the only integer eigenvalues of the chain under consideration. This fact suggests that this chain can be regarded as a finite-dimensional analog of the class of
quasi-exactly solvable Schrdinger operators, which has been extensively studied in the last two decades.
We have applied the method of moments to study some statistical properties of the chains spectrum, showing in particular that the density of eigenvalues follows a Wigner-like law. Finally, we emphasize that, unlike
the original freezing trick, the extension thereof developed in this paper can be applied to spin chains whose
associated dynamical spin model is only quasi-exactly solvable.
PACS: 75.10.Pq; 03.65.Fd
Keywords: Spin chains; Quasi-exact solvability; CalogeroSutherland models; Freezing trick
E-mail address: artemio@fis.ucm.es (A. Gonzlez-Lpez).

doi:10.1016/j.nuclphysb.2007.07.001
A. Enciso et al. / Nuclear Physics B 789 [FS] (2008) 452482
453
1. Introduction
Solvable spin chains have enjoyed a growing popularity in the last few years, due in part to
their novel applications to SUSY YangMills and string theories [16]. The prime example of
such chains is the celebrated Heisenberg model describing N spins in a one-dimensional lattice
with isotropic nearest-neighbors interactions independent of the site. The Hamiltonian of the
model is given by [7]

Si Si+1 ,
HHe =
(1)
i
y
where Si = (Six , Si , Siz ) is the spin operator of the ith site, the sum runs from 1 to N (as always hereafter), and SN+1 = S1 . As is well known, for spin 1/2 the model (1) can be exactly
solved using Bethes ansatz [810]. Several (partially) solvable generalizations of the Heisenberg chain (1) with short range interactions (at most between next to nearest neighbors) have
been subsequently proposed in the literature. These include, in particular, the family of chains
with arbitrary spin and nearest-neighbors interactions polynomial in Si Si+1 of Refs. [11,12], as
well as several models whose ground state can be written in terms of valence bonds [13,14].
A different type of solvable spin chain with long-range position-dependent couplings was
introduced independently by Haldane [15] and Shastry [16]. This chain describes a system of
N spins equally spaced on a circle, such that the strength of the interaction between each pair
of spins is inversely proportional to their chord distance. The motivation for introducing the HS
chain (2) was the fact that its exact ground state coincides with Gutzwillers variational wave
function for the Hubbard model [1719] when the strength of the on-site interaction tends to
infinity. We shall write the HaldaneShastry (HS) Hamiltonian as
1
i
sin(i j )2 (1 Sij ), i
,
HHS =
(2)
2
N
i<j
where Sij is the operator exchanging the ith and j th spins. Although the particles spin in the
original HS chain was assumed to be 1/2, one can more generally consider particles with n
internal degrees of freedom transforming under the fundamental representation of su(n). In this
case the spin permutation operators can be written in terms of the fundamental su(n) generators
Jk at each site k (normalized so that tr(Jk Jk ) = 12 ) as

n
1
1
Ji Jj .
+2
n
2
Sij =
=1
Note that for spin 1/2 particles (n = 2), we have Si = (Ji1 , Ji2 , Ji3 ).
The HS chain is naturally related to the scalar Sutherland model of AN type [20,21] and its
spin version introduced in Refs. [2224]. In fact, Polychronakos [25] noted that the complete
integrability of the HS chain could be deduced from that of the spin Sutherland model by suitably taking the strong coupling limit (the so-called freezing trick). Moreover, the latter author
applied this technique to construct an integrable spin chain related to the Calogero (rational)
model of AN type [26]. The Hamiltonian of this chain, usually referred to in the literature as the
PolychronakosFrahm (PF) chain, reads

(i j )2 (1 Sij ),
HPF =
(3)
i<j
454
where the chain sites i are now the equilibrium positions of N particles in the scalar part of the
potential of the Calogero spin model of AN type. In particular, the sites of the PF chain are not
equally spaced, unlike those of the HS chain. Indeed, it can be shown that the sites of the PF
chain are the zeros of the N th Hermite polynomial [27,28], which satisfy the system of algebraic
equations
1
, i = 1, . . . , N.
i =
(4)
i j
j =i
It turns out that both the HS and the PF chains, featuring long-range position-dependent interactions, can be solved in a more detailed and explicit way than the chains of Heisenberg type,
characterized by the short range and position independence of the interactions. For instance, the
partition function of the models (2) and (3) can be evaluated in closed form for arbitrary N using
the freezing trick [29,30], and its expression is relatively simple in both cases. The spectrum,
which consists of a set of integers (consecutive in the case of the PF chain), is highly degenerate
due to an underlying Yangian symmetry [31]. Moreover, it has been recently shown that even
for moderately large N the density of eigenvalues of the HS chain is Gaussian to a high degree
of approximation, and that the density of spacings between consecutive levels follows a simple
distribution different from the usual Poisson or Wigner laws [30]. In fact, there is strong evidence
that these results also hold for other spin chains of HaldaneShastry type, as, e.g., the BCN or
the supersymmetric versions of the original HS chain [32,33].
In this paper we shall consider a novel type of spin chain, which in some sense is intermediate
between the Heisenberg and the PolychronakosFrahm chains. Its Hamiltonian is obtained from
that of the PF chain (3)(4) by retaining only nearest-neighbors interactions, namely

(i i+1 )2 (1 Si,i+1 ),
H=
(5)
i
where the sites i are defined by restricting the sum in Eq. (4) to nearest neighbors:
i =
1
1
+
,
i i1 i i+1
i = 1, . . . , N.
(6)
In the previous equations we are identifying SN,N +1 with SN1 , N+1 with 1 and 0 with N .
We shall see that the chain (5) possesses several remarkable properties, whose study is the
purpose of this paper. In the first place, the spin chain (5) is related along the lines of the freezing
trick to the spin dynamical model
H =
x2i + a 2 r 2 +

i

2a 2
2a
(a Si,i+1 ), (7)
+
(xi xi1 )(xi xi+1 )
(xi xi+1 )2
i
where r 2 = i xi2 , a > 1/2, and we have identified x0 xN and xN+1 x1 . We have shown in
our recent papers [34,35] that an infinite proper subset of the spectrum of the Hamiltonian (7)
can be computed in closed form, so that this model is quasi-exactly solvable (QES) [3638]. The
Hamiltonian (7) is a spin version of the QES scalar model
Hsc = H |Si,i+1 1
introduced in [39] by Auberson et al. In particular, it was shown in the latter reference that
a 2
= e 2 r
|xi xi+1 |a
i
(8)
(9)
455
is the ground state function of the model (8), with eigenvalue E0 = N a(2a + 1). We shall prove
that the sites i of the chain (5) are in fact the coordinates of the unique maximum of in the
domain

C = x R N x1 < < xN .
(10)
By numerically solving Eq. (6), we shall see that for sufficiently large N the chain sites are normally distributed with zero mean and unit variance. We shall present a simple deduction of this
property based on the analysis of the continuous limit of the algebraic system (6) as N . As
a byproduct, we shall obtain an analytic formula providing a very accurate approximation to the
sites coordinates, valid even for moderately large values of N . Another nontrivial consequence
of this formula is the fact that for large N the strength of the coupling between the spins i and
i + 1 as a function of their mean coordinate (i + i+1 )/2 also follows the Gaussian law, but with
zero mean and variance 1/2.
The spectral properties of the spin chain Hamiltonian (5) are also remarkable. Indeed, by a
suitable modification of the freezing trick one can show that H possesses the eigenvalues 0, 1, 2
for arbitrary values of N and n, and exactly compute their corresponding eigenstates. Our numerical simulations evidence that these energies are the three lowest ones, and that for spin 1/2
none of the remaining eigenvalues of H are integers.1 For n > 2, the spectrum of H also contains the integer eigenvalue 3, which appears to be not the fourth but the fifth lowest energy. The
above properties suggest that the model (5) could be regarded as a quasi-exactly solvable chain,
in the sense that only a certain number of eigenvalues and their corresponding eigenvectors of
the Hamiltonian H can be computed in closed form.
We have also studied the distribution of energy levels of the chain (5) for a large number of
particles. Since the partition function of this chain is not known, we have performed a numerical
calculation of the density of levels using the methods of moments [40,41]. It turns out that,
in contrast with the typical behavior of spin chains of HaldaneShastry type, the distribution of
levels clearly deviates from the Gaussian law. As an indication of the accuracy of the approximate
level density derived via the moments method, we have compared its mean and variance with
the exact values obtained by taking traces of suitable powers of the Hamiltonian. From this
discussion it also follows that for large N the mean and variance of the energy behave as N 3 and
N 5 , respectively, just as for the trigonometric chains of HS type in Refs. [30,32].
The paper is organized as follows. In Section 2 we study the distributions of the chain sites
and the couplings, comparing the results obtained with those for the PF chain. Section 3, which
is the core of the paper, is devoted to the determination of the integer eigenvalues of the chain (5)
and their corresponding eigenstates. We also present in this section a detailed example for the
case of 5 particles of spin 1, which motivates a number of conjectures on the degeneracy of the
integer levels. In Section 4 we use the method of moments to approximately compute the density
of levels of the chain H, showing that it follows a Wigner-like law. Finally, in Section 5 we summarize our conclusions and outline possible future developments. For the readers convenience,
in Appendix A we present some background material on the exact eigenfunctions of the spin dynamical model (7) used in Section 3, while in Appendix B we include an overview of the method
of moments.
1 As a matter of fact, the previous assertion does not hold for the cases N = 3 (for which the chain (5) reduces to the
PF chain, whose eigenvalues are known to be integers), and N = 4 (for which our numerical simulations indicate that
all the eigenvalues are also integers). Therefore, in the rest of the paper we shall exclude these special cases from our
discussion.
456
2. The chain sites

We shall start this section by proving that the sites of the chain (5) are the coordinates of
the unique maximum in the domain (10) of the ground state function (9) of the scalar Hamiltonian (8). It is convenient to write the ground state as
= ea(x) ,
where
(x) =
log |xi xi+1 |
r2
2
has the same extrema as and is independent of a. Thus the equations (6) defining the chain
sites are just the conditions for = (1 , . . . , N ) to be a critical point of . The existence of a
maximum of in C is clear, since it is continuous in C and tends to both on its boundary
and as r . Uniqueness follows from the fact that the Hessian of is negative definite in C.
Indeed, by Gerschgorins theorem [42, 15.814], the eigenvalues of the Hessian of at x lie in the
union of the intervals

2
2
2

,
+
=
,
where
i
i
i
x x .
x 2
x 2
i
j =i
Since
2
= 1 (xi xi+1 )2 (xi xi1 )2 ,
xi2
2
= (xi xi1 )2 ,
xi xi1
2
= 0 if j = i, i 1,
xi xj
we have
2
+ i = 1,
xi2
and thus all the eigenvalues of the Hessian of are strictly negative.
Since and C are invariant under the transformation
xi xNi+1 ,
i = 1, . . . , N,
and has a unique maximum in C, it follows that

i = Ni+1 ,
(11)
so that the chain sites are symmetric about the origin. In particular, the center of mass of the spins
vanishes, i.e.,
1
i = 0.

(12)
N
i
We have numerically solved equations (6) for the chain sites for up to 250 spins. Before
presenting our conclusions, a remark on the configuration of these sites is in order. The attentive
reader might have been surprised by our claim that the chain (5) features only nearest-neighbors
457
Fig. 1. Sites of the chain H for N = 20 spins.
interactions, in spite of the fact that the first spin interacts with the last one. We can avoid this
objection by regarding the site coordinate i as an arc length in a circle of radius 2N / , see
Fig. 1. In this way the spins at the sites 1 and N are indeed nearest-neighbors and, moreover, the
strength of the interactions are inversely proportional to the squared distance between consecutive
spins, measured along the arc.
It is apparent from Fig. 1 (and also follows immediately from Eq. (6)) that the sites i are
not equally spaced. In fact, our computations show that for large values of N the sites i follow
with great accuracy a Gaussian distribution with zero mean and unit variance. More exactly, the
cumulative density of sites (normalized to unity)

F (x) = N 1
(13)
(x i ),
i
where is Heavisides step function, is approximately given by

F (x) =
1 + erf x/ 2 .
2
(14)
The agreement between the functions F and F is remarkably good for N 100 (see Fig. 2 for
the case N = 150) and increases steadily with N , e.g., the mean square error of the fit for 100,
150 and 200 spins are respectively 2.6 105 , 1.1 105 and 7.9 106 .
The fact that for large N the cumulative density of sites is well approximated by the Gaussian
law (14) can be justified by the following heuristic argument. Let x(t, u) be a smooth function
such that x(i, N ) = i for i = 1, . . . , N , and define the rescaled function y(s, ) = x(s/ , 1/ ).
By Eq. (6), the latter function must satisfy the relation
1
1
+
= y(s, )
y(s, ) y(s , ) y(s, ) y(s + , )
for = 1/N
1 and s =
all
1. Writing
y(s, ) =

k=0
1 2
N , N , . . . , 1.
yk (s) k ,
(15)
Let us now assume that Eq. (15) holds for all s R and
458
Fig. 2. Cumulative distribution functions F (x) (at its discontinuity points) and F (x) (continuous grey line) for N = 150
spins.
and using the expansion

y(s, ) y(s
, ) = y0 (s)

y0 (s)

y1 (s) 2 + O 3
2
the leading term in Eq. (15) yields the differential equation

y0 = y0 y0 2 .
The general solution of this equation is implicitly given by

s = c0 + c1 erf y0 (s)/ 2 .
(16)
Hence, up to terms of order = 1/N , the cumulative distribution function of the chain sites
(normalized to unity) is approximated by the continuous function

F (x) = c0 + c1 erf x/ 2 .
The normalization conditions F () = 0 and F () = 1 imply that c0 = c1 = 1/2, and thus
the empiric law (14) is recovered.
From Eq. (16) (with c0 = c1 = 1/2) it follows that the site k can be determined up to terms
of order 1/N by the formula

2k N
.
k 2 erf1
(17)
N
If the sites k were exactly given by the previous formula, they would satisfy the identity
2

erf k / 2 + erf Nk+1 / 2 = ,
N
which is clearly inconsistent with the exact relation (11). However, the slightly modified formula

1 2k N 1
k 2 erf
(18)
N
459
Fig. 3. Sites coordinates k and their continuous approximation (18) for N = 20 spins.
differs from (17) by a term of order 1/N and is fully consistent with the relation (11). Although
both (17) and (18) provide an excellent approximation to the chain sites for large N , the latter
equation is always more accurate than the former, and can be used to estimate k with remarkable
precision even for relatively low values of N , cf., Fig. 3.
It is also of interest to determine whether the position of the last spin tends to infinity as
N , since according to our interpretation of the chains geometry the number 2N / is the
radius of the circle on which the spins lie. From Eq. (18) it follows that for large N the last spins
coordinate N is approximately given by

1
1
N 2 erf
(19)
,
1
N
so that N should diverge as N . Of course, this assertion should be taken with some caution, since in Eq. (18) the argument of the inverse error function is correct only up to terms of
order 1/N . In order to check the correctness of the approximation (19), we use the asymptotic
expansion of erf1 (u) for u 1 in Ref. [43] to replace (19) by the simpler formula

N 2 log ,
(20)
where

N
= log .
As can be seen in Fig. 4, the approximate formula (20) qualitatively reproduces the growth of N
when N ranges from 100 to 250. A greater accuracy can be achieved by introducing an adjustable
parameter in Eq. (19) through the replacement 1/N /N , so that in Eq. (20) becomes

N
= log .
(21)

In Fig. 4, we have also plotted the law (20)(21) with the optimal value = 0.94, which is in
excellent agreement with the numerical values of N for N = 100, 105, . . . , 250.
460
Fig. 4. Position of the last spin N for N = 100, 105, . . . , 250 and its continuous approximation (20)(21) for = 0.94
(solid line) and = 1 (dashed line).
The last property of the spin chain (5) that we shall analyze in this section is the dependence
of the coupling between neighboring spins on their mean coordinate. Calling
k + k+1
k =
,
2
we shall now see that when N 100 the Gaussian law
hk = (k k+1 )2 ,
(22)
N 2 2
(23)
e k
2
holds with remarkable precision, cf. Fig. 5. Indeed, if x = x(k) denotes the RHS of Eq. (18) we
have

x
2k = N erf + N + 1,
2
hk
so that
2 1 x 2
dx
=
e2
dk
N
is of order 1/N . Hence, up to terms of order 1/N we have

2

1
dx
N 2 x 2 (k)
1 2
x k

=
.
x k+
e
2
2
dk
2
Since (up to terms of order 1/N 2 )

1
1
1
x(k) x k
+x k+
,
2
2
2
Eq. (23) follows from (25) replacing k by k + 12 .
(24)
(25)
461
Fig. 5. Plot of the coupling between the spins k and k + 1 versus their mean position, fitted by the Gaussian (23), for
N = 100 spins.
We shall finish this section with a brief comparison of the previous properties with those of
the PF chain (3). For large N , the density of sites of the PF chain (normalized to unity), that is the
density of zeros of the N th Hermite polynomial, is asymptotically given by the circular law [44]
N (x) =
1
2N x 2 .
N
(26)
site N of the
The last site N of the PF chain grows with N much faster than the corresponding
chain (5), for the largest zero of the N th Hermite polynomial behaves as 2N + O(N 1/6 ); see,
e.g., [45]. Finally, in contrast with the PF chain, the polynomials determined for each N 2 by
the corresponding sites i of the chain (5) do not form an orthogonal family. In other words, the
polynomials

p (N ) (z) (z i )
(27)
i
do not satisfy a three-term recursion relation of the form

p (N+1) (z) = zp (N ) (z) aN p (N1) (z),
as it can be verified using the explicit expressions
p (2) (z) = z2 1,
p (3) (z) = z3
3z
,
2
1
p (4) (z) = z4 2z2 + .
4
3. Quasi-exact solvability of the spin chain

In this section we shall see that the spin chain (5) and the spin dynamical model (7) are related
via a slight modification of the usual freezing trick mechanism. We shall exploit this connection
to compute in closed form some eigenstates of H with integer energy, for any number of particles
462
and arbitrary spin. We shall then compare our results with those obtained by numerical diagonalization of H for small values of N and n. The numerical computations strongly suggest that the
exact states derived in this section exhaust all the eigenstates of the chain (5) with integer energy.
We shall use in what follows the decomposition
H = Hsc + 2ah(x),
where
h(x) =
(xi xi+1 )2 (1 Si,i+1 ).
(28)
(29)
It shall also be convenient to write the scalar Hamiltonian (8) as

Hsc =
x2i + a 2 U (x) aV (x),
(30)
where the scalar potentials U and V are respectively given by

2
2
+
,
U (x) = r 2 +
(xi xi1 )(xi xi+1 )
(xi xi+1 )2
i
i

2
.
V (x) =
(xi xi+1 )2
(31)
(32)
From Eq. (30) it follows that for large a the low-lying eigenfunctions of Hsc concentrate at
absolute minima of the potential U in the domain (10) and their energies satisfy [46,47]
E = a 2 min U (x) + O(a).
xC
By comparison with the exact results for the ground state (9), one concludes that is an absolute
minimum of U and U ( ) = 2N . This condition has been used in the previous section to check
the accuracy of the numerical solution of the sites equations (6) for large values of N .
There are two main limitations which prevent the use of the standard freezing trick argument
[29,32] for computing the spectrum of the spin chain (5). The first one is the requirement that
be the unique minimum of the potential U in the domain C. Although our numerical calculations
suggest that this is indeed the case, we have not been able to provide a rigorous proof of this
fact. The second limitation, which is more fundamental, is the fact that the dynamical models H
and Hsc are only quasi-exactly solvable. Let us briefly recall the basics of the usual freezing trick
method in order to understand why the full knowledge of the spectra of H and Hsc is essential
for its application. Indeed, if the potential U has an unique minimum in C, for sufficiently
large a all the eigenfunctions of Hsc are sharply peaked around this point. Thus, if (x) is an
eigenfunction of Hsc with energy Esc and | is an eigenstate of the chain H with eigenvalue E,
for a 1 we have
h(x)(x)| (x)h( )| (x)H| = E(x)| .
By Eq. (28), the state (x)| is then an approximate eigenfunction of H with eigenvalue
E Esc + 2aE.
(33)
In other words, the Hamiltonian (7) is approximately diagonal in a basis of the form {i (x)|j },
where {i (x)} is a basis of eigenfunctions of Hsc and {|j } is a basis of eigenstates of H.
463
Eq. (33) cannot be used directly to compute the corresponding spectrum of H, since it is not clear
a priori which pairs of eigenvalues of H and Hsc yield an approximate eigenvalue (E Esc )/(2a)
of H. However, using Eq. (33) one can easily express the partition function Z of the spin chain
H in terms of the partition functions Z and Zsc of H and Hsc , respectively, via the formula
Z(2aT )
(34)
.
Zsc (2aT )
The latter formula, which is the key result behind the standard freezing trick approach, cannot be
used to compute the spectrum of the spin chain H unless the whole spectrum of both H and Hsc
is known.
In spite of the above limitations, we shall see in this section that it is still possible to compute a
number of eigenstates of the spin chain (5) from some of the families of spin eigenfunctions of the
model (7) constructed in Ref. [35]. More precisely, we will show that certain linear combinations
of these eigenfunctions factorize as the product of the ground state of the scalar model (8) times
a spin function, whose limit as a is an eigenstate of the spin chain Hamiltonian H.
Let us begin by introducing some preliminary notation. Let be the space of internal degrees
of freedom of N particles with su(n) spin, and denote the elements of its canonical basis by
|s1 . . . sN , where si = M, M + 1, . . . , M and M = (n 1)/2. Let be the total symmetrizer
under particle permutations, that is
Z(T ) = lim
1
k ,
N!
N!
k=1
where k denotes a permutation operator acting simultaneously on spatial coordinates

and spins.
We shall consider in what follows the subspace of spin states |s such that i |si,i+1 is
symmetric, where |sij is defined by

x1 x2 |s =
(35)
xi xj |sij ;
i<j
see Ref. [35] for a complete characterization of this subspace. Finally, given a state |s we
define the spin functions

(k) x; |s = x1k |s (k = 0, 1, 2),
(36)
(2) x; |s = x1 x2 |s ,

(3)
(37)
x; |s = x1 x2 (x1 x2 )|s .
We shall suppress one or both of the arguments of the above spin functions when appropriate.
We shall next present the basic result we have used to construct eigenstates of the chain (5) out
of the eigenfunctions of the dynamical spin model (7) in Appendix A. Recall that the energies of
these eigenfunctions are the numbers Elm = E0 + 2a(2l + m), where l and m are non-negative
integers and E0 is the ground state energy of the scalar model Hsc .
The key point in the ensuing argument is the fact that for large a the normalized ground state
0 = / of Hsc , where is given by (9) and 2 = C 2 , is sharply peaked around its
maximum . Hence, if F (x) is a continuous spin-dependent function such that the integral of
20 F over C is finite, the main contribution to this integral comes from a small ball centered at ,
up to exponentially small terms. By the standard argument behind the proof of Laplaces method,
it follows that

lim
(38)
20 F = F ( ) lim
20 = F ( ).
a
464
Assume now that (x;a) = 0 (x; a)(x; a) is a continuous eigenfunction of H with energy
Elm such that (x; a) = ki=0 a i i (x), with each i a polynomial in x. Denoting by

(f, F ) = f (x)F (x) dN x
C
the usual spin-valued inner product of a scalar function f (x) with a spin-valued function F (x),
from Eq. (28) we obtain

Elm 20 , = Elm (0 , ) = (0 , H ) = (0 , Hsc ) + 2a(0 , h ).
(39)
Since Hsc is self-adjoint and h is a matrix multiplication operator, the RHS of the previous
equation equals

(Hsc 0 , 0 ) + 2a 20 , h = E0 (0 , 0 ) + 2a 20 , h = 20 , (E0 + 2ah) .
From Eq. (39) it follows that
k

a i 20 , (h 2l m)i .
0 = 20 , (h 2l m) =
i=0
Taking the limit a , and using Eq. (38) and the fact that H = h( ), we finally obtain
(H 2l m)0 ( ) = 0.
(40)
0 ( ) = lim ( ; a)
(41)
Thus
a
is either zero or an eigenstate of the chain (5) with integer energy 2l + m.

We shall now see that the application of the previous method to the eigenfunctions (A.1)
(A.5) of the dynamical model (7) listed in Appendix A yields the following types of eigenstates
of the chain (5):

0 |s = (0) (|s) |s,
(42a)

(1)
1 |s = ; |s ,
(42b)

2 |s = (2) ; |s + (N 1) (2) ; |s ,
(42c)

(3)
(1)
3 |s = ; |s + 2 ; |s .

(42d)
For the states (42c) |s must belong to the subspace and be symmetric in the first two spins,
whereas for the states of type (42d) |s should be antisymmetric under S12 . In all cases, the
energy of the states k (|s) is the integer k. We shall see in what follows that the states (42d) of
energy 3 only appear for spin M > 1/2 (n > 2).
(0)
Note, first of all, that Eq. (A.8) implies that 2 = N , since otherwise each eigenfunction l0
(cf. Eq. (A.1)) would yield an eigenstate of the chain (5) with eigenvalue 2l for arbitrary l. (The
fact that 2 = N can also be established directly from Eq. (6) using the identity U ( ) = 2N
for the scalar potential (31).) Hence we need only consider the eigenfunctions (A.1)(A.5) with
(0)
(2)
l = 0. Moreover, from Eqs. (A.1)(A.5) and (A.9) it follows that the eigenfunctions 0m , 0m
(2)
(1)
(3)
and 0m (respectively 0m and 0m ) vanish identically at x = for odd (respectively even) m.
465
For the first type of eigenfunctions (A.1), when m is even Eq. (A.10) implies that
m
m
(0)
(0)
0m (x) a 2 1
0 0m a
N2
m x m (0) ,
2 !
where (0) is defined in (36). By Eq. (12), the RHS of the previous equation vanishes at x =
unless m = 0. Thus the eigenfunctions (A.1) only yield the zero energy eigenstates (42a) of H
(when l = m = 0). Note that the fact that any symmetric spin state is an eigenstate of the chain
H with zero energy follows directly from Eq. (5). Conversely, the latter equation implies that the
ground state energy of H is zero and that the corresponding eigenstates are totally symmetric.
For this reason we shall concentrate in what follows on the nontrivial states (42b)(42d).
Let us next examine the eigenfunctions of type (A.2) for l = 0 and odd m 1. In this case we
have
m1

1m
N 2
(1)
(1)
m1 x m1 (1)
0m (x) a 2 1
0 0m a
( 2 )!

x
(0) ,
(43)
(1)
where (1) is given by (36). Using again Eq. (12) we conclude that lima 0m vanishes at
x = unless m = 1. In this case, we have

(1)
lim 01 ( ) = (1) ; |s 1 |s .
a
(2)
(2)
Consider now the eigenfunctions 0m and 0m with even m. Using again Eq. (A.10) we
have
m

m
2N 2 x m
(2)
(0) + O a 1 ,
a 2 1
(44a)
0
0m
m2
(m 1)( 2 )!
m
a 2 1
0 0m =
m
2N 2 x m
(2)
(N
1)(m 1)( m2
2 )!

(0) + O a 1 ,
(44b)
where m 2. Thus the O(1) term of the left-hand sides of Eq. (44) vanish at x = for all m 2,
(2)
(2)
on account of Eq. (12). However, if 0m and 0m are built from the same spin state |s, the
m
(2)
previous equations imply that the O(1) part of the linear combination a 2 1
0 (0m + (N
(2)
1) 0m
) vanishes. This observation suggests considering the function
(2)

m
(2)
(2)
0m (x) a 1 2 1
0 0m + (N 1)0m
=
m
2 1
( m2
2 )!

(0) + O a 1 ,
x m2 (2) + (N 1) (2) N x 2 (1) x
where m 2 is even and (2) , (2) are given in Eq. (36). Note that 0m is only defined for states
(2)
(2)
|s symmetric under S12 , since otherwise 0m would not be defined. The limit of 0m as
a vanishes at x = unless m = 2, in which case we obtain the eigenstate with energy 2

(2)
lim 02 ( ) = (2) ; |s + (N 1) (2) ; |s 2 |s .
(2)
Let us finally turn to the last type of eigenfunctions (A.5) with l = 0 and odd m 3. From
Eq. (A.10) it immediately follows that

m1

1m
4N 2 x m1
x (0)
(1)
(3) =
+ O a 1 ,
a 2 1
(45)
0
0m
m3
m
(m 2)( 2 )!
466
whose limit as a vanishes identically at x = since m 3 in this case. However, as for the
(3)
previous states, one can cancel the leading term in a of 0m with a suitable linear combination
(0)
(1)
(3)
of 0m and 0m all built from the same spin state (necessarily antisymmetric under S12 for 0m
to be defined). Indeed, using Eqs. (43) and (45) and taking into account that for odd m
a
1m
2
1
0 0m =
(0)
m1
2
( m1
2 )!

x m (0) + O a 1 ,
one can easily check that

(3)
0m (x) a
3m
2

2(m 1) (1) 2(m 1)2 (0)
(3)
1
= O(1).
0
0m
m 2 0m
m(m 2) 0m
(46)
(3)
We shall next see that the O(1) term of 0m (x) vanishes at x = unless m = 3. Indeed, Eq. (A.9)
(+i,)
implies that for k > 2k the term x k Pk
(t) vanishes at x = to all orders in a. Hence the
(3)
O(1) terms of 0m (x) and

3m
m 1 2 (+1,)
(+3,)
m3
(3)
(1)
2
x
(t) 2
(t)
a
(47)
x P m1
P m3
m2
2
2
coincide at x = . From Eq. (A.9) it is straightforward to show that the O(1) term of (47) is a
linear combination of x m3 and x m1 , thus establishing our claim.
By the previous remarks, we need only compute the O(1) term of (47) for m = 3, which by
Eq. (A.10) is given by

2 2
(3) +
r N (N + 2)x 2 (1) .
N
Using Eq. (12) and the identity 2 = N we finally obtain the eigenstates (42d) of energy 3:

(3)
lim 03 ( ) = (3) ; |s + 2 (1) ; |s 3 |s .
a
The previous discussion guarantees that the nontrivial states k (|s) given by Eqs. (42b)(42d)
are eigenstates of the chain H with energy k provided that they do not vanish. For instance, the
states (42b) are easily seen to vanish when the spin state |s is symmetric, since in this case

1 |s = (1) ; |s = |s = 0.
This is also the case for the states (42c). Indeed, if |s is symmetric it clearly belongs to and
we have

r2
2
xi xj |s = N x 2 |s.
(2) x; |s + (N 1) (2) x; |s = |s +
N
N
i<j
Since 2 (|s) coincides with the LHS of this expression evaluated at x = , it vanishes on account
of Eq. (12). Less trivially, let |s be a linear combination of basic states |s |s1 , . . . , sN such
that |s| 2, where |s| denotes the number of distinct components of s (s1 , . . . , sN ). If, in
addition, |s is antisymmetric under S12 , we shall prove below that
2

r (1)
(2) = 0,
x
(3) + 2
(48)
N
467
which implies that 3 (|s) = 0 by Eq. (12) and the identity 2 = N . In summary, the state |s in
the definition of the states (42b)(42d) can be taken without loss of generality as follows:

1 |s : |s (),
(49)

S12 |s = |s,
2 |s : |s (),
(50)

3 |s : |s =
(51)
cs |s,
S12 |s = |s.
|s|3
Note, in particular, that the states (42d) do not appear for spin 1/2 in view of the first condition (51).
Our next task is to study the number of linearly independent states of the form (42b)(42d)
with |s satisfying the above conditions. The main difficulty in this respect is the fact that the
eigenstates k (|s) constructed from a set of linearly independent states |s satisfying conditions
(49)(51) need not be independent. In order to address this problem, it is convenient to introduce
some additional notation. Given a basic state |s with |s| = p, we define its spin content as the
set of pairs (s i , i ), where s 1 < < s p are the distinct components of s and i is the number of
times that s i appears in s. For instance, the spin content of the basic state |s = |0, 2, 1, 2, 1 is
{(2, 2), (0, 1), (1, 2)}. We shall say that an arbitrary spin state |s has well-defined spin content
if it is a linear combination of basic states having the same spin content. It is obvious that a
set of states whose spin contents are all different is linearly independent. Note also that if |s
has a well-defined spin content, then k (|s) has the same spin content. Therefore, it suffices
to determine the number of linearly independent states k (|s) built from states |s with welldefined spin content. Clearly, for a given spin content {(s i , i )} this number is independent of the
particular values of the spin coordinates s i . We shall therefore denote by dk (1 , . . . , p ) dk ()
the dimension of the linear space of states k (|s) with a given spin content {(s i , i )}. We shall
next prove the following upper bounds on these dimensions:

p1
.
d2 () p 1,
d3 ()
d1 () p 1,
(52)
2
Consider in the first place the states of the form 1 (|s) with a given spin content. If |s and |s
are two basic states with the same spin content and differing by a permutation of the last N 1
spin coordinates, then (x1 |s) = (x1 |s ), which implies that 1 (|s) = 1 (|s ) by Eqs. (36)
and (42b). Hence, the space of states 1 (|s) with spin content {(s 1 , 1 ), . . . , (s p , p )} is spanned
by the states

1 s i . . . , i = 1, . . . , p,
(53)
where the ellipsis denotes any ordering of the remaining N 1 spin components corresponding
to the above spin content. On the other hand, the p states (53) satisfy the linear relation
p

i 1 s i . . . = 0,
(54)
i=1
which implies the first inequality in (52). Indeed, if |s is a basic state with spin content {(s i , i )},
we have
N x|s
p
N

xi |s =
i x1 s i . . . .
i=1
i=1
468
Setting x = and using (12) we immediately obtain (54).

Let us turn now to the states of the form 2 (|s), where |s satisfies (50) and has a well-defined
spin content {(s 1 , 1 ), . . . , (s p , p )}. The results of our previous paper [35] (see Proposition 4)
imply that the space of states of this form is spanned by

2 |i ,
i = 1, . . . , p,
(55)
where

j (k j k )s j s k . . . , i > 1,
i (i 1)s i s i . . .
1j,kp
j,k=i
|i =

i j j i
j s s . . . + s s . . . , i = 1.
1j p
(56)
j =i
However, from [35, Proposition 3] it follows that the p states (55) satisfy the linear relation
p

2 |i = 0,
(57)
i=1
which yields the second inequality in (52).

Consider finally the states of the form 3 (|s), where |s is antisymmetric under S12 and
has a certain spin content {(s 1 , 1 ), . . . , (s p , p )}. Clearly, the space of such states is spanned
by

3 s i s j . . . s j s i . . . , 1 i < j p.
(58)
p
We shall prove below that the latter 2 states satisfy the linear relations
p

j 3 s i s j . . . s j s i . . . = 0,
i = 1, . . . , p.
(59)
j =1
Multiplying the LHS of the ith relation by i and summing over i we obtain
p

i j 3 s i s j . . . s j s i . . . ,
i,j =1
which vanishes by antisymmetry. Thus we can safely drop one of the identities (59), say the last
one, and the first p 1 are independent since they
can be solved
for 3 (|s i s p . . . |s p s i . . .)
p1
p
for i = 1, . . . , p 1. Hence there are at most 2 (p 1) = 2 independent states of the
form (58), which establishes the last inequality in (52).
We still have to prove the identities (59). To this end, denote by 3 3 (x; |s) the LHS of
Eq. (48). Since 3 ( ; |s) = 3 (|s), Eq. (59) follow directly from the identities
p

j =1

j 3 x; s i s j . . . s j s i . . . = 0,
i = 1, . . . , p.
(60)
469
Table 1
Energies and degeneracies of the first six and the last
levels of the chain (5) with N = 5 and M = 1
Energy
Degeneracy
0
1
2
2.035
3
3.953
15.033
21
24
24
15
6
15
In order to establish (60), note first that

N

(2) s i s j . . . =
x1 xl (x1 xl )s i s j . . .
r 2 (1) s i s j . . . N x
l=1
p

k (3) s i s k . . . ,
k=1
so that
p

N
j 3 s i s j . . . s j s i . . .
2
j =1
=N
p

j =1
p
p

(3) i j
(3) i k
s s ...
s s ... +
j
j k
j k (3) s j s k . . . .
j,k=1
j,k=1
(61)
p
Since j =1 j = N , the first two terms of the RHS of the previous formula cancel, whereas the
last one vanishes by antisymmetry. This concludes the proof of Eq. (60). Note, finally, that the
latter equation for p = 2 easily yields the identity (48).
3.1. Example
We shall now discuss in detail the case N = 5 and spin M = 1. The sites of this chain are given
by 5 = 1 = 1.469, 4 = 2 = 0.584, 3 = 0. The Hamiltonian (5) is represented by a 243
243 real symmetric matrix, whose numerical diagonalization is straightforward. The spectrum
consists of 21 different levels, the only integer energies being 0, 1, 2, 3. For conciseness sake,
we only present the first six levels and the last one with their corresponding degeneracies, cf.
Table 1. Notice, in particular, the appearance of a noninteger energy between the third and the
fifth levels. The degeneracy of the ground level is given by the number of independent symmetric
states, that is, the number of combinations with repetitions of 5 elements from 3. Turning to the
other integer energies, we shall next present a basis of eigenstates of each nontrivial type (42b)
(42d). By the previous discussion, we must first determine all the possible spin contents {(s i , i )}
compatible with conditions (49)(51), and then construct a basis of states k (|s) with each of
these spin contents. In practice, the former task is performed in two steps, namely, one first finds
all possible degeneracy vectors , and then one determines the spin contents associated with each
of these vectors.
470
Table 2
List of basic states |s generating the states (42b) of unit energy for each
degeneracy vector
|s
(4, 1)
(3, 2)
(2, 3)
(1, 4)
(3, 1, 1)
(2, 2, 1)
(2, 1, 2)
(1, 3, 1)
(1, 2, 2)
(1, 1, 3)
|0, |+, |0000+

|00, |++, |000++
|000, |+++, |00+++
|0000, |++++, |0++++
|0+, |0+
|00+, |00+
|0++, |0++
|000+, |000+
|00++, |00++
|0+++, |0+++
Table 3
States | of the form (56) generating the states (55) of energy 2 for
each degeneracy vector
|
(4, 1)
(3, 2)
|0, |+, |0000+

3|00 |00, 3|++ |++,
3|000++ |++000
3|000 |000, 3|+++ |+++,
3|+++00 |00+++
|0000, |++++, |++++0
6|0+ |0+ |+0,
3(|0+ + |0+) + |0+ + |+0
|00+ |00+ |0+0 |+00,
|00+ |00+ |+00 |+00
|0++ |++0 |0++ |+0+,
|++0 |++0 |0++ |0++
3(|000+ + |000+) + |+000 + |+000,
6|000+ |+000 |+000
|00++ |++00 |++00 |+00+,
|++00 |00++ |00++ |00++
|0+++ + |0+++ + 3(|+++0 + |+0++),
6|+++0 |0+++ |0+++
(2, 3)
(1, 4)
(3, 1, 1)
(2, 2, 1)
(2, 1, 2)
(1, 3, 1)
(1, 2, 2)
(1, 1, 3)
The states (42b) of unit energy are generated by basic states |s with p 2 different spin
components. In Table 2 we present the list of such basic states, where we have taken into account
the relation (54) to drop one state for each different spin content; for instance, the state |0
does not appear in the table, since 1 (|0) = 41 (|0). In this way we obtain 24
basic states, whose corresponding states 1 (|s) turn out to be linearly independent.
Similarly, the space of states 2 (|s) of energy 2 is spanned by the states (55) with |i given
by (56), for each spin content {(s 1 , 1 ), . . . , (s p , p )} such that p 2. As in the previous case,
the relation (57) implies that for each spin content one can drop one of the states |i . In Table 3
we present a possible choice of the 24 states |i that can be constructed by this procedure. Just
as before, the corresponding states 2 (|i ) are easily seen to be linearly independent.
Finally, for spin 1 the set of states 3 (|s) of energy 3 is spanned by states of the form (58)
associated with spin contents with three different components, cf. Eq. (51). By the last inequality
471
Table 4
List of states |s generating the states (42d) of energy 3 and their corresponding degeneracy vectors
|s
(3, 1, 1)
(2, 2, 1)
(2, 1, 2)
(1, 3, 1)
(1, 2, 2)
(1, 1, 3)
|0+ |+0
|+00 |+00
|0++ |0++
|+000 |+000
|00++ |00++
|0+++ |0+++
(52), for each such spin content there is at most one independent state of the form (58). Thus
there are at most 6 independent states of the form (42d), generated (for instance) by the states |s
listed in Table 4. As in the previous cases, it can be checked that the 6 states 3 (|s) constructed
from the states in Table 4 are actually independent.
Several important remarks can be made in connection with the previous example. In the first
place, it is apparent that the inequalities (52) are in this case equalities for every spin content. Secondly, the number of independent states with integer energy k of the form k (|s) coincides with
the degeneracy of the corresponding level, cf. Table 1. Hence, in this example all the eigenstates
with integer energy are of the form (42), and can therefore be computed explicitly. In fact, we
have performed a similar study for the case of N = 6 and spin M = 3/2, arriving at exactly the
same conclusions. In view of these examples, it is natural to formulate the following conjectures:
(1) The inequalities (52) are always equalities.
(2) The only integer energies of the chain (5) are 0, 1, 2 and (for M 1) 3, corresponding to the
first three and the fifth levels.
(3) The only eigenstates with integer energy are those of the form (42).
According to the first conjecture, for k = 1, 2, 3 the number of independent states k (|s) with
a well-defined spin content with p elements depends only on p, and is given by

p 1, k = 1, 2,
dk (p) = p1
2 , k = 3.
By the third conjecture, the degeneracy of the integer energy k is thus
dk =
min(n,N)

p=1

n
N 1
dk (p)
,
p
p1
k = 1, 2, 3.
(62)

Indeed, there are pn different choices of p distinct spin values s i , and N1
p1 ways of selecting
p numbers i 1 such that 1 + + p = N . In order to evaluate the sum in (62), note first that
min(n,N)

p=1
p1
j

min(n,N)
n N j 1
n
N 1
N 1
=
.
p
p1
j
p
pj 1
On the other hand, expanding both sides of the identity

n
1 + z1 (1 + z)Nj 1 = zn (1 + z)N+nj 1
p=1
(63)
472
in powers of z, we easily obtain

min(n,N)

p=1
n
p

N j 1
pj 1

=

N +nj 1
.
N
The previous identity and Eq. (63) immediately yield the following explicit formula for the degeneracy of the positive integer energies of the chain (5):

N 1
N + n k k1
, k = 1, 2, 3.
dk =
(64)
k 1 + k1
N
(The degeneracy
of the ground level E = 0 is the dimension of the space of symmetric states,

.)
namely N+n1
N
We have numerically diagonalized the chain Hamiltonian (5) for several values of n and N
such that nN 38 = 6561. In all cases, we have checked the validity of the second conjecture,
and that the formula (64) for the degeneracies of the integer energies (which is a direct consequence of the first and third conjectures) is satisfied. These results lend strong numerical support
to the above conjectures, whose rigorous proof deserves further investigation.
4. Statistical analysis of the spectrum
It has been recently shown that for large N the level density of spin chains of HS type can be
approximated with remarkable accuracy by a Gaussian distribution [30,32,33]. A natural question is whether this is also the case for the spin chain (5) under consideration. The fact that the
partition function of the chain (5) has not been computed in closed formunlike those of the HS
chains in the previous referencesmakes it difficult to address this problem directly. Although
one can in principle diagonalize numerically the matrix of the chain Hamiltonian H, in practice
this is only feasible for relatively small values of N . In this section we shall study the level density of the chain (5) using the method of moments outlined in Appendix B, which provides an
accurate estimation of the level density for larger values of N .
In order to improve the numerical stability of the method, it is convenient to take the parameters b and c in Eq. (B.11) as
b = 0,
c = 1,
(65)
so that the spectrum of A lies in the interval [2, 2]. Therefore the matrix A and the Hamiltonian H are related by
A=
4H
2,
Emax
(66)
where Emax is the largest eigenvalue of H. Note that this eigenvalue can be computed numerically
without difficulty for relatively large values of N , since the matrix of H is very sparse. One must
then check that the coefficients bk and ck in the continued fraction expansion of the resolvent
of A (cf. Appendix B) approximately stabilize to the values b and c in Eq. (65) for k0 < k < k1 ,
with k1 k0 . We have verified that this is indeed the case for N sufficiently large, where k0 is
typically of the order of 10. We can thus use Eq. (B.12) to obtain a continuous approximation
g(x), where x (4E/Emax ) 2, to the density of eigenvalues of the matrix A, as explained in
Appendix B.
473
Fig. 6. Continuous approximation g(x) to the density of eigenvalues of the matrix (66) for spin 1/2 and N = 12
(Emax = 201.21) compared with the histogram of the spectrum of this matrix.
As a test of the accuracy of the method, we have computed the density g for spin 1/2 and
N = 12, since in this case the matrix A can still be diagonalized numerically. In Fig. 6 we have
compared the approximate density g computed by applying the moments method with k0 = 20
and 20 random vectors with the histogram of the spectrum of A obtained by subdividing the
interval [2, 2] in 50 subintervals. As can be seen from the latter figure, the continuous density
g essentially reproduces the shape of the histogram.
The real interest of the method of moments is the possibility of approximating the density of
eigenvalues of a large matrix whose numerical diagonalization is not feasible. We have been able
to compute the approximate density g(x) of eigenvalues of the matrix A for N up to 21 for spin
1/2, and up to 13 for spin 1. For instance, in Fig. 7 we present the plot of g for spin 1/2 and
N = 21, computed with k0 = 20 and 20 random vectors. It is apparent from this plot, and for
similar plots for spins 1/2 and 1, that the level density is not Gaussian. As a matter of fact, the
function g is well approximated by the Wigner-like distribution
f (x) = N 1 y e
)
(y
y = x + 2,
(67)
where the normalization constant N is given by

+1
2
1 +1
2
1 3
1
1
, ;
+ 2
,
N = 2
1 F1
1 F1 , ;
2
2
2
2
2
2
2 2
and 1 F1 is the confluent hypergeometric function of the first kind. Thus the behavior of the chain
H is rather different in this respect from that of the spin chains of HS type.
When N is greater than 12, the accuracy of the level density obtained by the method of
moments cannot be gauged directly by computing the spectrum of the matrix A. An indirect way
of estimating this accuracy consists in comparing the first few moments of the density g with
those of the spectrum of A, computed by taking the traces of appropriate powers of H. As an
example, we shall next derive simple expressions for the mean and variance of the spectrum of
H in terms of finite sums involving only the coordinates of the chain sites, which can be easily
474
Fig. 7. Approximate density g(x) of eigenvalues of the matrix (66) (grey line) for spin 1/2 and N = 21 (Emax = 1126.53),
compared with the distribution (67) with optimal parameters = 0.85, = 1.27, = 0.91 (dashed line).
Fig. 8. Logarithmic plot of the relative errors 1 (black boxes) and 2 (crosses) in Eq. (70) as a function of the number
of sites N for spin 1/2. The approximate level density g in Eq. (71) has been computed taking k0 = 20 and averaging
over 20 random vectors.
approximated for very large values of N . We shall see in this way that the agreement between
these values and those obtained from g is indeed very good and, roughly speaking, improves as
N grows.
In the first place, the mean energy E of H can be easily computed noting that tr Si,i+1 = nN1 ,
so that (cf. Eq. (22))

1
hi nN nN1 = 1
hi .
E = nN tr H = nN
(68)
n
i
475
Similarly, since

hi hj (1 Si,i+1 Sj,j +1 + Si,i+1 Sj,j +1 )
H2 =
i,j
and
tr(Si,i+1 Sj,j +1 ) = nN2+2ij ,
a straightforward calculation yields

2

2
N2
hi + (n + 1)
h2i ,
tr H = (n 1)n
(n 1)
i
so that the variance of the energy is given by

2
1 2
2
N
2
hi .
n tr H E = 1 2
n
(69)
In Fig. 8 we present a logarithmic plot of the relative errors

1
|E Eg |
,
E
2
| 2 g2 |
(70)
between the exact values (68)(69) and their approximations

Emax
Eg =
4
2
(x + 2)g(x) dx,
g2
E2
= max
16
2
(x + 2)2 g(x) dx Eg2
(71)
for N = 12, . . . , 21 and spin 1/2. From the latter plot it is apparent that for N 15 both errors
are less than 0.1%, which suggests that for large N the continuous function g(x) computed by
the method of moments is indeed an excellent approximation to the level density of the matrix
A. From Eqs. (68) and (69) and the discussion of the distribution of the chain sites in Section 2 it
is straightforward to deduce the asymptotic behavior of the mean and the variance of the energy
for large N . Indeed, using Eq. (24) we easily obtain
2
2
dx
N 2 2
hk (k+1 k )2 x(k + 1) x(k)
(72)

e k,
dk
2
so that

k

3 2
N 2 2 N 2 2
N k2
N3
hk
e k
e k (k+1 k ) e 2
e 2 x dx
3/2
2
2
(2)
2
k
N3
.
2 3
Hence for large N the mean energy E behaves as

N3
1
E
.
1
n
2 3
(73)
476
Table 5
A comparison of some properties of the spin chain H versus the Heisenberg chain HHe and the PolychronakosFrahm
chain HPF (the properties followed by an asterisk are based on unpublished work by the authors)
Chain sites distribution
Solvability
Integer energies
Level density
Mean energy growth
Energy variance growth
HHe
HPF
Gaussian
Quasi-exact
First few
Wigner-like law (67)
N3
N5
Equispaced
Exacta
No
Wigner-like
N
N
Circular law (26)

Exact
All
Gaussian
N2
3
N
a Only for spin 1/2.
Similarly,

k
h2k

5 2
N 4 2 2
N k2
N5
N5

e k (k+1 k ) e 2
e 2 x dx =
,
2
5/2
4
(2)
2
4 2 5
k
and therefore the large N limit of the variance of the energy is given by

1
N5
2

(74)
1 2 .
n
4 2 5
Thus, the leading behavior of the energy mean and variance of the chain (5) is analogous to
that of the trigonometric HS spin chains studied in Refs. [30,32]. In particular, both quantities
grow with N much faster than their counterparts for the Heisenberg chain (both of which diverge
as N ).
5. Conclusions
In this paper we have studied a novel spin chain with position-dependent nearest neighbors
interactions, which is intermediate between the Heisenberg chain (position-independent, nearest
neighbors interactions) and the spin chains of HaldaneShastry type (position-dependent, longrange interactions); see Table 5 for a brief comparison.
We have developed a new method related to Polychronakoss freezing trick which has made
it possible to compute in closed form a certain number of energy levels with their corresponding
eigenstates, for any number of particles and arbitrary spin. While the eigenvalues of the original
HaldaneShastry and PolychronakosFrahm chains (the closest HS analogs of the chain under
consideration) are all integers, the only integer eigenvalues of the new chain are precisely those
that have been computed exactly by our method. This fact strongly suggests that the chain (5)
should be regarded as the first instance of a quasi-exactly solvable spin chain, thus extending
the usual notion of quasi-exact solvability to finite-dimensional Hamiltonians. It is interesting to
note, in this respect, that the three lowest frequencies of the small oscillations near the equilibrium of the classical potential (31) are also integers [48].
The method developed in this paper is quite general, and only relies on the explicit knowledge
of a number of eigenfunctions of the corresponding spin dynamical model. In this respect, it goes
one step beyond the usual freezing trick, which requires the computation in closed form of the
whole spectrum of the related dynamical model. In particular, our method is well-suited to spin
chains whose associated dynamical model is only quasi-exactly solvable, like for instance the
models with elliptic interactions constructed in Refs. [49,50].
477
Acknowledgements
This work was partially supported by the DGI under grant no. FIS2005-00752, and by the
Complutense University and the DGUI under grant No. GR69/06-910556. A.E. acknowledges
the financial support of the Spanish Ministry of Education and Science through an FPU scholarship. The authors would also like to thank V. Martn-Mayor for useful discussions on the
moments method.
Appendix A. Exact eigenfunctions of the dynamical spin model (7)
In this appendix we shall list the eigenfunctions of the dynamical spin model (7) used to
construct the eigenstates of the chain (5) presented in Section 3. Let

1
3
1
2r 2
=N a+
1,
,
(m) = 1 m N a +
,
t=
2
2
2
N x 2

where x = N1 i xi is the center of mass coordinate. In Ref. [35] it was shown that the model
(7) possesses the following families of spin eigenfunctions with energy Elm = E0 + 2a(2l + m),
where l 0 and m is as indicated in each case:

(,)
ar 2 P m (t) (0) , m 0,
2
2 (+1,) (1)

(1)
m1
Ll ar P m1 (t) x
(0) , m 1,
lm = 0 x
2

(+2,)

(2)
lm
= 0 x m2 Ll ar 2 P m 1 (t) (2) 2x
(1) + x 2 (0)
(0)
lm
= 0 x m Ll
2( + 1)
2 m1
2 +1
(A.1)
(A.2)
(+1,)
P m 1 (t) (0)
2
m 2,
(A.3)

2 (+2,) (2)

(2)
m2
Ll ar P m 1 (t) 2x
(1) + x 2 (0)
lm = 0 x
2
2( + 1)
(+1,)
P m 1 (t) (0)
2
, m 2,
x
(2 m1
2 + 1)(N 1)

2 (+3,)
2 3 (0)
(3)
m3
(3)
(2)
2 (1)
lm = 0 x
Ll ar P m3 (t) 2x
+ 2x x
2
3
+

2x 2
(+2,)
(t) 2(
P
m

2 2 1 m3
2
+ 2) (1) x
(0)

2(2 + 3)
(+1,)
(t) (0) ,
(m)x 3 P m3
m(m 2)
2
(A.4)
m 3.
(A.5)
In the previous formulas denotes the integer part and (m) = 12 (1 (1)m ) is the parity of m.
The spin functions (k) , (2) , (3) (cf. Eqs. (36) and (37)) are built from a state |s symmetric
(2)
(3)
under S12 and belonging to for lm , and antisymmetric under S12 for lm . The generalized
( ,)
Laguerre polynomials Ll and the Jacobi polynomials Pk

appearing in Eqs. (A.1)(A.5) are
478
defined as
Ll (z) =

l

l zj
(1)j
,
l j j!
(A.6)
j =0
1 1
(k)j ( + + k + 1)j ( + j + 1)kj (1 z)j ,
k!
2j j !
k
( ,)
Pk
(z) =
(A.7)
j =0
where
(x)j = x(x + 1) (x + j 1)
is the Pochhammer symbol. The eigenfunctions of type (A.5) are independent of the remaining
ones (A.1)(A.4) only for spin greater than 1/2 (n > 2), cf. [35]. The model (7) possesses two
additional families of eigenfunctions (also derived in the previous reference), which have not
been listed above as they do not yield any further eigenstates of the chain (5).
(+i,)
Let us now discuss the behavior of the terms Ll (ar 2 ) and x 2k Pk
(t) in Eqs. (A.1)(A.4)
as a . Consider first the polynomial Ll (ar 2 ). Since

(N a)lj
l
=
1 + O a 1 ,
lj
(l j )!
it follows that

a l Ll
(N r 2 )l

ar 2 =
+ O a 1 ,
l!
(A.8)
where the term O(a 1 ) in the previous equation is actually a polynomial in a 1 . (Throughout
this paper, the symbol O(a k ) denotes any function f (a) such that a k f (a) has a finite (possibly
zero) limit as a .) For the other type of term, note that
(+i,)
x 2k Pk
(t) =
k
1 (k)j
( + + i + k + 1)j
k!
Nj j!
j =0
( + i + j + 1)kj x
2(kj )
N x r
2

2 j

(A.9)
is clearly a polynomial in x. Taking into account that

( + i + j + 1)kj = (N a)kj 1 + O a 1 ,
we obtain
a k x 2k Pk
(+i,)
(t) =

N k 2k
x + O a 1 ,
k!
(A.10)
where O(a 1 ) is polynomial in a 1 and x.

Appendix B. The method of moments
The method of moments [40,41] is a powerful tool for computing the density of eigenvalues
of a large Hermitian matrix whose spectrum is not known explicitly. The method is based on the
479
relation between a probability distribution g(x), i.e., a nonnegative function whose integral over
its support [x1 , x2 ] is one, and its resolvent
x2
R(z) =
x1
g(x)
dx,
zx
Im z = 0.
Using the well-known identity

= (x),
lim
0+ 2 + x 2
it is straightforward to show that
1
(B.1)
lim R(x i ), x (x1 , x2 ).
0+
The previous formula makes it possible to compute the probability density g(x) if the resolvent
R(z) is known. One of the key ingredients of the method is the fact that if g is positive in a set
of nonzero measure, the resolvent can be expanded as a continued fraction
g(x) =
R(z) =
1
,
z b0 zbc11
(B.2)
where bk , ck are the coefficients in the three-term recursion relation

Pk+1 (x) = (x bk )Pk (x) ck Pk1 (x)
(P1 = 0, P0 = 1, c0 = 1)
(B.3)
satisfied by the orthogonal polynomial system {Pk (x)}

k=0 associated with the density g(x). The
polynomials Pk are the monic polynomials determined by the orthogonality condition
x2
Pk Pl
Pk (x)Pl (x)g(x) dx = 0,
k = l.
x1
It can be shown [51] that these polynomials satisfy a three-term recursion relation of the
form (B.3) with coefficients bk and ck given by
bk =
xPk2
Pk2
ck =
Pk2
2
Pk1
(B.4)
In this paper we are interested in computing the density of eigenvalues (normalized to unity)
of a Hermitian d d matrix A, i.e.,
1
(x i ),
d
d
g(x) =
i=1
where 1 d are the eigenvalues of A. The corresponding resolvent

1
1
(z i )1 = tr(z A)1
d
d
d
R(z) =
i=1
is thus a rational function of z. In this case the resolvent can also be expanded as a continued
fraction of the form (B.2) terminating at level d, i.e., cd = 0 and Eqs. (B.3) and (B.4) hold for
k d 1.
480
If the density g is not known, the coefficients bk and ck cannot be computed from Eq. (B.4),
and hence the expansion (B.2) cannot be used directly to evaluate the resolvent. On the other
hand, the coefficients bk and ck can be determined once the moments x k of the distribution
g are known. Another central idea of the method consists in replacing x k by the average of
the expectation values (v, Ak v) over a suitable set of normalized random vectors v Rd . More
precisely, let ui , i = 1, . . . , d, be d independent random variables uniformly distributed in the
interval [1, 1], and let v be the vector with components vi = ui /u. Clearly, the components
of v satisfy
1
ij ,
d
where the bar stand for the average over the random numbers ui . We thus have
vi vj =
(B.5)
d
d
d

k

1 k
1 k 1 k
k
v, A v =
A ij vi vj =
A ii = tr A =
i
d
d
d
i,j =1
1
d
i=1
d
i=1

x k (x i ) dx = x k .
(B.6)
i=1
The previous equality implies that

P (x) = v, P (A)v ,
where P is an arbitrary polynomial. From Eq. (B.4) we immediately obtain the following formula
for the coefficients bk and ck :
bk =
(wk , Awk )
wk 2
ck =
wk 2
wk1 2
(B.7)
where wk Pk (A)v. By Eq. (B.3), the vectors wk satisfy the recursion relation
wk+1 = (A bk )wk ck wk1
(w1 = 0, w0 = v, c0 = 1).
The latter formula, together with Eq. (B.7), can be used to recursively compute the coefficients
bk and ck for all k = 1, . . . , d 1.
The number d = nN is typically very large, so that it is not feasible to compute all the coefficients bk and ck by the procedure just outlined. In practice, one only computes the first few of
these coefficients, say up to k = k0 , and tries to estimate the part Tk0 (z) of the continued fraction (B.2) involving the remaining coefficients. In many cases of interest [52], the coefficients
bk and ck approximately stabilize for a wide range of values of k, namely bk b and ck c
for k0 < k < k1 < d, where k1 k0 . In this case the remainder Tk0 (z) can be approximated as
follows:
c
Tk0 (z)
T (z).
c
z b zb
But the truncation factor T (z) satisfies
T (z) =
c
,
z b T (z)
481
so that

1
z b (z b)2 4c .
2
We thus obtain the following approximate formula for the resolvent:
T (z) =
R(z)
1
z b0
c1
xb1 .
(B.8)
(B.9)
ck
0
zbk T (z)
0
The previous equation can be written as
A(z) iB(z) (z)

,
R(z)
C(z) iD(z) (z)

with A, B, C, D polynomials in z with real coefficients and
(z) = (z x1 )(x2 z), x1,2 = b 2 c;
(B.10)
(B.11)
note that x1,2 R, on account of Eq. (B.7). The RHS of (B.10) has a branch cut on the interval
[x1 , x2 ], which is the support of its corresponding density function. From Eqs. (B.1) and (B.10)
we finally obtain the following approximate formula for the normalized density of eigenvalues
of the Hermitian matrix A:
|A(x)D(x) B(x)C(x)| (x)

, x [x1 , x2 ].
g(x)
(B.12)
C(x)2 + D(x)2 (x)

Since the RHS of the previous equation cannot vanish identically in any subinterval of [x1 , x2 ]
with nonzero length, it follows that the formula just obtained is only appropriate when the spectrum of A has no gaps.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
J.A. Minahan, K. Zarembo, JHEP 0303 (2003) 013.

D. Berenstein, S.A. Cherkis, Nucl. Phys. B 702 (2004) 49.
R. Roiban, A. Volovich, JHEP 0409 (2004) 032.
N. Beisert, M. Staudacher, Nucl. Phys. B 727 (2005) 1.
L. Freyhult, C. Kristjansen, T. Mnsson, JHEP 0512 (2005) 008.
A.S. Gorsky, Theor. Math. Phys. 142 (2005) 153.
W. Heisenberg, Z. Phys. 49 (1928) 619.
H. Bethe, Z. Phys. 71 (1931) 205.
L. Hulthn, Ark. Mat. Astron. Fys. 26A (1938) 1.
J. des Cloizeaux, J.J. Pearson, Phys. Rev. 128 (1962) 2131.
H.M. Babujian, Phys. Lett. A 90 (1982) 479.
L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
C.K. Majumdar, D.K. Ghosh, J. Math. Phys. 10 (1969) 1388.
I. Affleck, T. Kennedy, E.H. Lieb, H. Tasaki, Phys. Rev. Lett. 59 (1987) 799.
F.D.M. Haldane, Phys. Rev. Lett. 60 (1988) 635.
B.S. Shastry, Phys. Rev. Lett. 60 (1988) 639.
J. Hubbard, Proc. R. Soc. London, Ser. A 276 (1963) 238.
M.C. Gutzwiller, Phys. Rev. Lett. 10 (1963) 159.
F. Gebhard, D. Vollhardt, Phys. Rev. Lett. 59 (1987) 1472.
B. Sutherland, Phys. Rev. A 4 (1971) 2019.
B. Sutherland, Phys. Rev. A 5 (1972) 1372.
482
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
Z.N.C. Ha, F.D.M. Haldane, Phys. Rev. B 46 (1992) 9359.

K. Hikami, M. Wadati, J. Phys. Soc. Jpn. 62 (1993) 469.
J.A. Minahan, A.P. Polychronakos, Phys. Lett. B 302 (1993) 265.
A.P. Polychronakos, Phys. Rev. Lett. 70 (1993) 2329.
F. Calogero, J. Math. Phys. 12 (1971) 419.
F. Calogero, Lett. Nuovo Cimento 20 (1977) 251.
H. Frahm, J. Phys. A: Math. Gen. 26 (1993) L473.
A.P. Polychronakos, Nucl. Phys. B 419 (1994) 553.
F. Finkel, A. Gonzlez-Lpez, Phys. Rev. B 72 (2005) 174411.
D. Bernard, M. Gaudin, F.D.M. Haldane, V. Pasquier, J. Phys. A: Math. Gen. 26 (1993) 5219.
A. Enciso, F. Finkel, A. Gonzlez-Lpez, M.A. Rodrguez, Nucl. Phys. B 707 (2005) 553.
B. Basu-Mallick, N. Bondyopadhaya, Nucl. Phys. B 757 (2006) 280.
A. Enciso, F. Finkel, A. Gonzlez-Lpez, M.A. Rodrguez, Phys. Lett. B 605 (2005) 214.
A. Enciso, F. Finkel, A. Gonzlez-Lpez, M.A. Rodrguez, J. Phys. A: Math. Theor. 40 (2007) 1857.
A.V. Turbiner, Commun. Math. Phys. 118 (1988) 467.
M.A. Shifman, Int. J. Mod. Phys. A 4 (1989) 2897.
A.G. Ushveridze, Quasi-Exactly Solvable Models in Quantum Mechanics, Institute of Physics Publishing, Bristol,
1994.
G. Auberson, S.R. Jain, A. Khare, J. Phys. A: Math. Gen. 34 (2001) 695.
C. Benoit, E. Royer, G. Poussigue, J. Phys.: Condens. Matter 4 (1992) 3125.
J.L. Alonso, L.A. Fernndez, F. Guinea, V. Laliena, V. Martn-Mayor, Phys. Rev. B 63 (2001) 054411.
I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products, sixth ed., Academic Press, San Diego, 2000.
J.M. Blair, C.A. Edwards, J.H. Johnson, Math. Comp. 30 (1976) 827.
F. Calogero, A.M. Perelomov, Lett. Nuovo Cimento 23 (1978) 650.
D. Dominici, math.CA/0601078.
B. Simon, Ann. Inst. H. Poincar, Sect. A (N.S.) 38 (1983) 295.
D. Holcman, I. Kupka, Forum Math. 18 (2006) 445.
A. Khare, I. Loris, R. Sasaki, J. Phys. A: Math. Gen. 37 (2004) 1665.
F. Finkel, D. Gmez-Ullate, A. Gonzlez-Lpez, M.A. Rodrguez, R. Zhdanov, Commun. Math. Phys. 221 (2001)
477.
F. Finkel, D. Gmez-Ullate, A. Gonzlez-Lpez, M.A. Rodrguez, R. Zhdanov, Nucl. Phys. B 613 (2001) 472.
T.S. Chihara, An Introduction to Orthogonal Polynomials, Gordon and Breach, New York, 1978.
P. Turchi, F. Ducastelle, G. Trglia, J. Phys. C: Solid State Phys. 15 (1982) 2891.
A staggered six-vertex model with non-compact

continuum limit
Yacine Ikhlef a,b, , Jesper Jacobsen a,b , Hubert Saleur b,c
a LPTMS, Universit Paris-Sud, Btiment 100, Orsay 91405, France
b Service de Physique Thorique, CEA Saclay, Gif-sur-Yvette 91191, France
c Department of Physics and Astronomy, University of Southern California, Los Angeles, CA 90089, USA
Received 8 March 2007; accepted 6 July 2007

Abstract
The antiferromagnetic critical point of the Potts model on the square lattice was identified by Baxter
[R.J. Baxter, Proc. R. Soc. London A 383 (1982) 43] as a staggered integrable six-vertex model. In this
work, we investigate the integrable structure of this model. It enables us to derive some new properties,
such as the Hamiltonian limit of the model, an equivalent vertex model, and the structure resulting from the
Z2 symmetry. Using this material, we discuss the low-energy spectrum, and relate it to geometrical excitations. We also compute the critical exponents by solving the Bethe equations for a large lattice width N.
The results confirm that the low-energy spectrum is a collection of continua with typical exponent gaps of
order (log N )2 .
1. Introduction
Our understanding of 1 + 1 conformal field theories with non-compact target spaces has improved a great deal in the last few years, thanks to the use of geometrical methods [2], and ideas
from string theory [3]. The topic is of the highest interest in the context of the AdS/CFT duality.
Theories with non-compact target spaces also play an important role in statistical mechanics.
A sophisticated example of this role occurs in the supersymmetric approach to phase transitions
for non-interacting disordered electronic systems, where the universality class of the transition
* Corresponding author at: LPTMS, Universit Paris-Sud, Btiment 100, Orsay 91405, France.
E-mail address: yacine.ikhlef@u-psud.fr (Y. Ikhlef).

doi:10.1016/j.nuclphysb.2007.07.004
484
Y. Ikhlef et al. / Nuclear Physics B 789 [FS] (2008) 483524
between plateaux of the integer quantum Hall effect is related with the IR limit of a non-compact
1 + 1 supersigma model at = [4]. A more basic example is provided by Brownian motion
and subtle properties thereof, such as the (non-)intersection exponents [5]. In both cases, the noncompacity of the target space occurs because the electron trajectories or the random path can visit
a given site (edge) an infinite number of times. This is in sharp contrast with self avoiding models
for which almost everything is by now understood, and related with ordinary CFTs (essentially,
a twisted free boson).
An obvious strategy to tackle the physics of models with non-compact target spaces is to start
with a lattice model having an infinity of degrees of freedom per site/link. For instance, it is easy
to generalize the usual XXX chain to a non-compact representation of SL(2, R), and try to use the
standard tools of Bethe ansatz, Baxter Q-operator, etc., to obtain properties such as gaplessness
and critical exponents. Despite some serious progress in this direction [6], the problem is far
from being closed.
Another strategy is based on the observation that a non-compact continuum limit may well
arise from a lattice model with finite number of degrees of freedom per site/link if the nonunitarity is strong enough. The two families of examples exhibited so far involve models with
supergroup symmetrieseither models with OSP(m|2n) symmetries (such as intersecting loop
models) [7] which, in their Goldstone phases can be described in the IR by a collection of free
bosons and symplectic fermions, or the SU(1|2) integrable spin chain with alternating 3 and 3
representations, which was found to be described by the SU(2|1) WZW model at k = 1 [8]. In
both cases, a continuous spectrum of critical exponents is found, and the target space does exhibit
some non-compact directions indeed.
The examples of Brownian motion and self intersecting dense curves should convince the
reader that non-compact target spaces are more common and useful than might have been surmised a few years ago. In a recent paper, we found [9] that the antiferromagnetic Potts model
on the square lattice for Q continuous has critical properties seemingly involving a twisted noncompact boson. This conclusion was based on some numerical and analytical evidence, and
implied that the well-known six-vertex model itself might exhibit such an exotic continuum limit
if properly staggered. The purpose of this paper is to discuss these results further, and put them
on considerably firmer grounds.
Indeed, the evidence for a continuous spectrum of critical exponents is not so easy to obtain
from studies of a finite lattice model. What was really established so far in [79] was that low energy levels appeared with extremely high degeneracies in the limit of long chains, and that naive
calculations of finite size corrections indicated truly infinite degeneracies. This wasthanks to
complementary arguments, such as mappings onto sigma models, or abstract construction of
WZW theories on supergroupsinterpreted as strong indications for a continuous spectrum in
the scaling limit. Direct evidence was however missing, together with estimates of the measure
of integration on the continuous spectrum, if any. These issues will be resolved here.
The paper is organized as follows. In Section 2, the critical antiferromagnetic Potts model and
the related staggered six-vertex model are defined. Symmetries and limiting cases are studied.
It is shown in particular that the model coincides with the OSP(2|2) model of [7] in one limit,
and the SO(4) model in another. In Section 3, the solution of the model using the Bethe ansatz
is discussed. In Section 4, a detailed analysis of the spectrum of conformal exponents from the
Bethe equations is carried out. Very accurate evidence for the existence of a continuous spectrum
is obtained, together with some information on the measure. This information is used in Section 5 to relate the results to theoretical expectations, in particular those of the supersphere sigma
485
model of [10]. Elements for a physical interpretation of the continuous spectrum are proposed in
Section 6. Conclusions are gathered in Section 7.
2. The staggered six-vertex model and its integrable structure
2.1. The general integrable six-vertex model
On the square lattice L of 2N vertical lines and M horizontal lines, associate the complex number v(J ) to the J th vertical line, and h(I ) to the I th horizontal line (see Fig. 1). The
parameters v(J ) and h(I ) are called line rapidities. On this lattice, define the general inhomogeneous six-vertex model with local weights given in terms of the difference uI J = h(I ) v(J ).
The weights that satisfy the YangBaxter equations are obtained by taking Eqs. (1.5)(1.6) of
Ref. [11] and performing the substitution:
i
t, (I ), (J ), (I, J ) ei , ei 2ih(I ) , e2iv(J ) , eih(I )iv(J )i ,
2
(I, J ), (I, J ), (I, J ) 1, 1, I J .
(1)
(2)
Thus, the weights of the inhomogeneous integrable six-vertex model (see Fig. 2) are:
1 (I, J ), . . . , 6 (I, J ) = sin( uI J ), sin( uI J ), sin uI J , sin uI J ,
I J eiuI J sin , (I J )1 eiuI J sin .
(3)
The parameters I J must satisfy the additional condition:

I,J +1 I +1,J = I,J I +1,J +1 .
Fig. 1. The spectral parameter uI J defined by the line rapidities.
Fig. 2. The allowed configurations of the six-vertex model.
(4)
486
The parameters I J do not alter the eigenvalues of the transfer matrix, thus they play the role of
gauge parameters. The parameter has the value:
= cos .
(5)
2.2. The staggered six-vertex model associated to the critical antiferromagnetic Potts model
The Potts model we consider is a model of interacting spins living on the vertices of the square
lattice L (see left-hand side of Fig. 3). Each spin si can take Q distinct values. It is coupled to
its nearest neighbours, with a coupling constant J1 (respectively J2 ) on the NW-SE (respectively
SW-NE) lattice edges. The partition function of the model is:

exp J1 (si , sj )
exp J2 (si , sj ) ,
ZPotts =
(6)
{si } ij 1
ij 2
where the sum is over all spin configurations, and each product corresponds to one type of edge
of L.
When the couplings J1 and J2 are negative, the model is antiferromagnetic. In this domain,
the critical line separating the paramagnetic and antiferromagnetic phases is given by the condition [1]:

J
e 1 + 1 eJ2 + 1 = 4 Q.
(7)
The model defined by Eq. (6) on the square lattice L can be mapped to a six-vertex model on
the square lattice L . This lattice is represented in full lines and black circles, and is called the
medial lattice of L (see Fig. 3). The weights of this six-vertex model depend on the parameters
J1 , J2 and Q. We use the notations:

(8)
Q = 2 cos ,
e J1 1
,
x1 =
Q
e J2 1
x2 =
.
Q
(9)
The equivalent six-vertex model has weights 1 , . . . , 6 on the even vertices and 1 , . . . , 6 on
the odd vertices, where:
1 , . . . , 6 = 1, 1, x1 , x1 , ei /2 + x1 ei /2 , ei /2 + x1 ei /2 ,
(10)
1 , . . . , 6
(11)
= x2 , x2 , 1, 1, e
i /2
+ x2 e
i /2
,e
i /2
+ x2 e
i /2
The parameter is independent of the parameters x1 and x2 :

=
1 2 + 3 4 5 6 1 2 + 3 4 5 6
=
= cos .
21 3
21 3
(12)
Fig. 3. The square lattice L (dotted lines and white circles) and its medial lattice L (full lines and black circles).
487
Fig. 4. The line rapidities corresponding to the antiferromagnetic critical point of the Potts model.
Fig. 5. The transfer matrix T (u). The value next to each vertex is the spectral parameter uI J .
The criticality condition (7) can be parametrized by:

x1 =
sin u
,
sin( u)
x2 =
cos( u)
.
cos u
(13)
When this condition is satisfied, the weights (10)(11) of the Potts model correspond to a
particular case of the integrable six-vertex model (3). The rapidity v(J ) is equal to 0 (respectively
/2) when J is even (respectively odd). The rapidity h(I ) is equal to u (respectively u + /2)
when I is even (respectively odd). This configuration of the line rapidities is shown in Fig. 4.
The gauge parameter is set to = ei /2 at every vertex. Thus, the partition function of the Potts
model at the antiferromagnetic critical point is described by the two-row transfer matrix of the
integrable six-vertex model with the above choice of the rapidities. We call this matrix T (u) (see
Fig. 5).
2.3. The R-matrix and the thirty-eight-vertex model
2.3.1. Building block: The R-matrix of the six-vertex model
Definition The R-matrix of the integrable six-vertex model is defined by its matrix element
R equal to the Boltzmann weight of the configuration shown in Fig. 6. Let V be the Hilbert
space generated by the vectors |, |. Then the R-matrix is a linear operator mapping the space
V V onto V V . The matrix elements of the R-matrix with spectral parameter uI J = u and
gauge parameter I J = are the Boltzmann weights (3). In the basis (|, |, |, |),
488
Fig. 6. The R matrix of the six-vertex model.
Fig. 7. The YangBaxter relation for the six-vertex R-matrix.
the R-matrix is:
sin( u)
0
R(u, ) =
0
0
eiu sin
sin u
0
0
sin u
1 eiu sin
0
0
0
.
0
sin( u)
(14)
Symmetries The R-matrix satisfies the YangBaxter equations shown in Fig. 7 and the inversion
relation:

R(u, )R u, 1 = sin( u) sin( + u)1.
(15)
It also has the symmetry property:
R(u + , ) = R(u, ).
(16)
The R-matrix preserves the total magnetization:

z + z = z + z .
(17)
Relation to the TemperleyLieb algebra When the gauge is set to = 1, the R-matrix can be
written in terms of the TemperleyLieb generator E:
R(u, 1) = sin( u)1 + sin uE,
where:
0
0
0 ei
E=
0
1
0
0
0
1
ei
0
0
0
.
0
0
(18)
(19)
489
Fig. 8. The matrix R(u), defining the antiferromagnetic critical point of the Potts model.
In the Hilbert space of row configurations V1 V2N , define the operators:

Em 11 1m1 E 1m+2 12N ,
m = 1, . . . , 2N,
(20)
with a non-trivial action on the space Vm Vm+1 . This family of operators satisfies the
TemperleyLieb algebra:

(Em )2 = QEm ,
(21)
Em = Em Em+1 Em = Em Em1 Em ,
Em Em = E m Em
if |m m | > 1.
(22)
(23)
The operators Em can be expressed in terms of the Pauli matrices:

1 x x
y y
z
z
1 i sin mz m+1
m m+1 + m m+1 cos mz m+1
2
or, in a more compact form:
Em =
+
+ m m+1
+
Em = m+ m+1
i z
1
z
m.
e
1 mz m+1
2
(24)
(25)
2.3.2. The R-matrix. Conservation laws

The six-vertex model defined in Section 2.2 is not homogeneous, and the transfer matrix T (u)
is built using R-matrices with different values of uI J . One can construct a homogeneous model
by considering the R-matrix, acting on double-edges. The double-edges live in the Hilbert space:
V V V.
(26)
The R-matrix acts on the product space V V.

As a consequence of the magnetization conservation by R, the R-matrix also preserves the
total magnetization:
z 1 + z 2 + z1 + z2 = z1 + z2 + z1 + z2 .
(27)
When the gauge parameter is set to = 1, another conserved quantity can be constructed. Start
from the operator c:
c (cos )1 R(/2, 1) = (cos )1 R(/2, 1).
(28)
490
Fig. 9. A graphical representation of the identity: (c c)1 R(c c) = R.
This operator can be expressed in terms of the TemperleyLieb generator:

c = 1 (cos )1 E.
(29)
According to the inversion relation (15), this operator has the property:
c2 = 1.
The R-matrix obeys the conservation rule:

R(u), c c = 0.
(30)
(31)
This is a consequence of the YangBaxter equations and the inversion relation, as shown in Fig. 9.
The eigenvectors of the operator c are:

|0 = (2 cos )1/2 ei /2 | ei /2 | ,
(32)

1/2 i /2
i /2
e
| + e
| ,
|0 = (2 cos )
(33)
|+ = |,
(34)
| = |.
(35)
The eigenspace associated to the eigenvalue 1 is {|+, |, |0}, and the eigenspace associated to
the eigenvalue 1 is {|0}.

2.3.3. Mapping to the 38-vertex model
The coefficients of the R-matrix in the basis:

|+, |, |0, |0
|+, |, |0, |0
(36)
define the Boltzmann weights of a 38-vertex model on the square lattice (see Fig. 10). In this
vertex model, each edge carries an arrow or a thick line. The state |+ (respectively |) is
represented by an up or right (respectively down or left) arrow. The state |0 is represented by an
by a thick line. Setting:
empty edge, and the state |0
0 = 2 ,
(37)
u0 = 2u,
(38)
a0 , b0 , c0 = sin(0 u0 ), sin u0 , sin 0
(39)
Fig. 10. The 38-vertex model.
the weights of the 38-vertex model read:

(1)
(8)
a1 = a1 =

1
(c0 )2 + a0 b0 ,
4
1
(2)
(4)
a1 = a1 = b0 c0 ,
4
1
(3)
(5)
a1 = a1 = a0 c0 ,
4
1
(6)
(7)
a1 = a1 = a0 b0 ,
4
1
(1)
(1)
a2 = a4 = e2iu a0 c0 ,
4
1
(2)
(2)
a2 = a4 = a0 c0 ,
4
1
(1)
(1)
a3 = a5 = e2iu a0 c0 ,
4
491
492
1
(2)
(2)
a3 = a5 = a0 c0 ,
4
1 i( 2u)
(1)
(1)
a6 = a8 = e
b0 c0 ,
4
1
(2)
(2)
a6 = a8 = ei b0 c0 ,
4
1
(1)
(1)
a7 = a9 = ei( 2u) b0 c0 ,
4
1 i
(2)
(2)
a7 = a9 = e b 0 c 0 ,
4
1
(1)
(1)
(1)
(1)
a10 = a11 = a12 = a13 = a0 b0 ,
4
1
(2)
(2)
(2)
(2)
a10 = a11 = a12 = a13 = a0 b0 ,
4
1
a14 = a15 = (a0 )2 ,
4
1
a16 = a17 = (b0 )2 ,
4
1 2iu
a18 = e
c0 (b0 a0 ),
4
1
a19 = e2iu c0 (b0 a0 ).
4
2.3.4. The limit /2
If the parameter is set to /2 and the spectral parameter u is fixed, then the weights of the
38-vertex model are equal to zero, except:
1 2
sin 2u,
4
1
(1,2)
(1,2)
(1,2)
(1,2)
a10 = a11 = a12 = a13 = sin2 2u,
4
1 2
a14 = a15 = a16 = a17 = sin 2u.
4
The R-matrix is proportional to the graded permutation P :
(1)
(8)
(6)
(7)
a1 = a1 = a1 = a1 =
1 2
sin 2uP ( /2, u fixed),
4
+, },
P |j |k = (1)gj gk |k|j , j, k {0, 0,
(41)
g0 , g0 , g+ , g = 0, 0, 1, 1.
(42)
R=
(40)
This trivial limit can be avoided by scaling the spectral parameter as:
=
+ ,
2
u=
+ w,
2
(43)
where 0 and w is fixed. In this rescaled limit, the Boltzmann weights are proportional
to 2 .
493
become degenerate, but the following combinations remain nonNote that the states |0, |0
degenerate:

= ei/4 tan(/2) 1/2 | + i| ,
|0 + i|0
(44)

= ei/4 tan(/2) 1/2 | i| .
|0 i|0
(45)
R/( 2 ) in the basis (|0, |0,
|+, |). One gets:
the matrix elements of R

a 1(1) = a 1(8) = 1 + w w 2 ,
(j )
Denote a i
a 1(2) = a 1(4) = w,
(3)
(5)
(1,2)
a 1 = a 1 = a 2
(1,2)
= a 3
(6)
(7)
(1,2)
(1,2)
a 1 = a 1 = a 10 = a 11
(1,2)
(1,2)
a 6
= a 8
= iw,
(1,2)
(1,2)
a 7
= a 9
= iw,
a 14 = a 15 = (1 w)2 ,
a 16 = a 17 = w 2 ,
(1,2)
(1,2)
= a 4
= a 5
= (1 w),
(1,2)
= a 12
(1,2)
= a 13
= w(1 w),
a 18 = a 19 = (1 2w).
These weights can be related to an integrable loop model with OSP(2|2) symmetry. Indeed, the
can be expressed as a combination of the identity, the permutation operator P defined
matrix R
in Eq. (41), and the TemperleyLieb operator E. The latter is defined in tensor notation as a
contraction of the spaces V , V and V , V (see Fig. 8):

E
(46)
= J J ,
|+, |):
where J is the bilinear form in the basis (|0, |0,
1 0 0 0
0 1 0 0
J =
J J = J J = 1.
,
0 0 0 i
0 0 i 0
(47)
By construction, the operator obeys the TemperleyLieb algebraic relations (21)(23), with:

E 2 = Tr J J E = 0.
(48)
|0,
|+ |, | |+}, the matrix of the operator E is:
In the block {|0 |0, |0
1
1 i
i
1 i
i
1
E=
.
i i 1 1
i
i
1 1
(49)
is equal to:
The matrix R
= (1 w)1 + wE + w(1 w)P .
R
(50)
These weights define an integrable loop model [12], with three allowed vertices, and a loop fugacity q = 0. The graphical correspondence is shown in Fig. 11. Note that according to Eq. (13),
494
Fig. 11. The three allowed vertices of the dense intersecting loop model.
the Potts model is isotropic when u = /2 + /4, which is equivalent to w = 1/2. This is also
the isotropic point of the loop model (50).
We conclude that, in the limit /2, the staggered six-vertex model coincides with the
OSP(2|2) integrable model. In this limit, the arrows represent the fermionic coordinates, and
the thin and thick lines the bosonic ones. The equivalence in the boundary conditions has to be
treated with some care however: the periodic vertex model corresponds to the antiperiodic sector
of the OSP system, with an effective central charge ceff = 1 + (2) 24 (1/8) = 2. This
means that the OSP symmetry is broken in the periodic vertex model, where the fermions are
twisted, i.e., they become a complex (Dirac) fermion instead of symplectic fermions.
2.3.5. The limit 0
The approach for this limit is similar to the case /2. Consider the following limit:
= ,
where
u = w,
0+
(51)
and w is fixed. It is convenient to perform the change of basis:
= i|0,
|0
= i|0,
|0
(52)
= |+,
|+
= |.
|
(54)
(53)
(55)
Define a transformation of the R-matrix that does not affect the partition function: the Boltzmann
weights of the vertices are multiplied by (1) for each /2-turn of the thick lines (the weight
(8)
a1 is not affected). Since the number of such turns is even for every thick polygon on the lattice,
the partition function is invariant under this transformation.
(j )
R/( 2 ) in the basis
Denote a i the matrix elements of the (rescaled) resulting matrix R
|+,
|0,
|).
One gets:
(|0,

a 1(1) = a 1(8) = 1 + w + w 2 ,
(2)
(4)
(3)
(5)
a 1 = a 1 = w,
(1,2)
a 1 = a 1 = a 2
(1,2)
= a 3
(6)
(7)
(1,2)
(1,2)
a 1 = a 1 = a 10 = a 11
(1,2)
(1,2)
= a 8
= w,
a 6
(1,2)
(1,2)
a 7
= a 9
= w,
a 14 = a 15 = (1 + w)2 ,
a 16 = a 17 = w 2 ,
(1,2)
(1,2)
= a 4
= a 5
= (1 + w),
(1,2)
= a 12
(1,2)
= a 13
= w(1 + w),
495
a 18 = a 19 = 1.
The corresponding loop model is constructed with two generators P and E. The permutation
operator P is defined as:
P = .
The TemperleyLieb operator E with parameter
E = J J ,
where
1
0
J =
0
0
0
1
0
0
0
0
0
1
(56)
Q = 4 is built using the simple contraction:
(57)
0
0
,
1
0
J 2 = 1.
(58)
= (1 + w)1 wE + w(1 + w)P .

R
(59)
is equal to:
The matrix R
These are the integrable weights of the loop model defined in [12] with a loop fugacity q = 4.
Note that, in this regime, the Potts model is anisotropic for any value of w. The loop model
itself is isotropic when w = 1/2: the relative weight for loop crossings is then equal to 1/2.
Following [12], this model is a graphical version of the SO(4) integrable model based on the
vector representation. Using that so(4) = sl(2) + sl(2) one can expect it to decouple into two
copies of the isotropic six-vertex model or XXX spin chain, a feature we will confirm when
discussing the associated Hamiltonian or Bethe equations.
2.4. Transfer matrices
In this section, we set the gauge to = 1, so that the transfer matrices can be related to the
generators of the TemperleyLieb algebra.
One-row transfer matrix The two-row transfer matrix T (u) is the product of two one-row
transfer matrices: T (u) = T (u + /2)T (u) (see Fig. 5). As a consequence of the YangBaxter
equations and the inversion relation, when periodic boundary conditions are imposed in the horizontal direction, the one-row transfer matrices commute:

T (u), T (u ) = 0.
(60)
The one-row transfer matrices have the property:
T (u + /2) = (1)N eiP T (u)eiP .
(61)
Define the operator eiP as the shift of all vertical edges by one site to the right. The operator c
defined by Eq. (28) is used to build a conserved quantity of the transfer matrix. Denoting cm the
charge operator acting on the space Vm Vm+1 , the global charge is:
C c1 c3 c2N 1 ,
(62)
C = 1.
(63)
496
Fig. 12. Diagrams defining the transfer matrices C (a), eiP C (b) and CeiP (c), for N = 4. The white circles represent
c operators.
If u = 0, the transfer matrices T (u), T (u + /2) become:

N

1
eiP C,
sin 2
T (0) =
2
N

1
CeiP .
T (/2) = sin 2
2
The matrices (62), (64), (65) are represented in Fig. 12.
(64)
(65)
Conservation laws for T As a consequence of the conservation of the total magnetization and
the charge c by the R-matrix, the transfer matrix T preserves the total magnetization:
1
z
S = 1z + + 2N
(66)
2
and the charge C defined above. Since T (u) also commutes with the transfer matrix T (0),
T (u) is invariant under the action of the one-site shift operator eiP .
2.5. Hamiltonian limit of the transfer matrix
To compute the derivative of the matrix T (u) at the points u = 0, u = /2, it is convenient to
write this matrix in terms of the R-matrix:

2N 2N1
...2N
1 2N
(u) =
R1211 (u)R2322 (u /2) R2N1
T11...
2N1 (u)R2N 2N (u /2).
2N
1 ,...,2N
Denote by the derivative with respect to u. According to Eq. (18):

R(u) = cos( u)1 + cos uE.
Differentiating the previous expression yields:
N1

N

1
cos2 t2j 1 + sin2 t2j ,
sin 2
T (0) =
2
j =1
(67)
(68)
497
Fig. 13. Diagrams defining t2j 1 and t2j of Eq. (68), for N = 4 and j = 3. The white circles represent c operators. The
arrow points to the vertex that differs from T (0).
where the matrices t2j 1 and t2j are defined by the diagrams on Fig. 13. Using the graphical
representation of the matrices T (0), t2j 1 , t2j in Figs. 1213 and the property (30), one gets the
intermediate results:

1

1
1
cV2j 3 V2j 2 cV2j 2 V2j 1 cV2j 3 V2j 2 ,
sin 2
T (0) t2j 1 =
(69)
2

1

1
1
cV2j 1 V2j .
sin 2
T (0) t2j =
(70)
2
The Hamiltonians H1 , H2 are defined as the logarithmic derivatives of the transfer matrix T (u)
at the points u = 0, /2:

1
1
sin 2 T (0) T (0)
2
N

=
cos 2 1 + cos (E2j 1 + E2j ) (E2j E2j 1 + E2j 1 E2j ) .
H1
(71)
j =1
Using Eq. (61),

1
1
sin 2 T (/2) T (/2)
2
= eiP H1 eiP
N

cos 2 1 + cos (E2j + E2j +1 ) (E2j +1 E2j + E2j E2j +1 ) .
=
H2
(72)
j =1
The Hamiltonian associated with the two-row transfer matrix T (u) is defined as:

1
1
sin 2 T (0) T (0).
2
According to the commutation relations (60),
H
H = H1 + H2
2N

cos 2 1 + 2 cos Em (Em+1 Em + Em Em+1 ) .
=
m=1
(73)
(74)
498
Note that the Hamiltonian H is the sum of two commuting parts:

H = H1 + H2 ,
[H1 , H2 ] = 0.
(75)
A consistent decomposition of the momentum operator is:

e2iP = eiP C CeiP = CeiP eiP C .
(76)
Indeed, the operators H1 , H2 , (eiP C), (CeiP ) all commute with one another.
Using the expression (25) of the TemperleyLieb generators, the Hamiltonian (74) is expressed in terms of Pauli matrices:
H=
2N

1
+
z
z
+ sin2 mz m+1
+ m m+2
mz m+2
m+ m+2
2
m=1

+
z
1
z
+
m m+1 + m m+1
cos 2 1 .
+ i sin m1
m+2
2
(77)
In the limit 0, the Hamiltonian (77) describes, as expected from the SO(4) identification,
two decoupled ferromagnetic XXX spin-chains:
H=
1
1
(
2j 1 2j +1 + 1)
(
2j 2j +2 + 1)
2
2
N
j =1
j =1
( = 0).
(78)
3. Bethe equations, Bethe states and eigenvalues

When periodic conditions are imposed in the horizontal direction, the staggered six-vertex
model is solvable by Bethe ansatz [11]. We call r the total number of particles.
3.1. Bethe equations for two types of particles
The Bethe ansatz equations are:
j {1, . . . , r}
r

exp 2iNk(j ) =
exp i(j , l ) .
(79)
l=1
The one-particle momentum and the scattering amplitude are given by:

sinh( + i )
exp 2ik() =
,
sinh( i )
(80)
sinh 12 ( 2i )

exp i(, ) =
.
sinh 12 ( + 2i )
(81)
The roots j describing the ground state and the physical excitations are expected to sit on the
two lines Im() = /2. Define two types of particles:
j+ = j + i/2,
j = 1, . . . , r+ ,
(82)
j = j i/2,
j = 1, . . . , r .
(83)
499
The one-particle momenta are:

cosh( + i )
exp 2i k()
exp 2ik i
=
.
2
cosh( i )
The scattering amplitude between two particles of the same type is:

sinh 12 ( 2i )
=
.
exp i1 (, ) exp i i , i
2
2
sinh 12 ( + 2i )
(84)
(85)
The scattering amplitude between two particles of different types is:

cosh 12 ( 2i )

=
.
exp i1 (, ) exp i i , i
2
2
cosh 12 ( + 2i )
Define the shifted scattering amplitudes as odd functions of ( ):

exp i1 (, ) = exp i1 (, ) ,

exp i1 (, ) = exp i1 (, ) .
(86)
(87)
(88)
The Bethe equations (79) split into two sets:

j {1, . . . , r+ }
j {1, . . . , r }
r+
e2iN k(j ) = (1)r+ 1
e2iN k(j ) = (1)r 1
ei1 (j ,l )
l=1
r+
r
ei1 (j ,l ) ,
(89)
l=1
r

i1 (j ,l )
i1 (j ,l )
(90)
1 (j , l ),
(91)
l=1
l=1
Take the logarithm of Eqs. (89)(90):

j {1, . . . , r+ }
j {1, . . . , r }
j ) = 2I +
2N k(
j
r+
j ) = 2I
2N k(
j
1 (j , l )
l=1
r+
r
l=1
r
1 (j , l )
l=1
1 (j , l ).
(92)
l=1
The Bethe integers Ij follow the following rules: if r+ (respectively r ) is even, then all
the Ij+ (respectively Ij ) are half-odd integers; if r+ (respectively r ) is odd, then all the Ij+
(respectively Ij ) are integers. Summing Eqs. (91)(92) over j and recalling that the functions
1 are odd, one relates the total momentum to the Bethe integers:
r+

r+
r
r

j) +
j) =
Ij +
Ij .
ktot =
(93)
k(
k(
N
j =1
j =1
j =1
j =1
The one-particle momenta and the scattering amplitudes can be written:

tan k()
= tanh tan ,

1
1 (, )
= tanh ( ) cotan ,
tan
2
2

1 (, )
1
tan
= tanh ( ) tan .
2
2
(94)
(95)
(96)
500
3.2. Bethe states

Form of the Bethe states The one-particle states are inhomogeneous plane-waves, represented in real space as:
1
(2m 1) = cosh ( i )e2ik()m ,
(97)
2
1
(2m) = sinh ( + i )e2ik()m
(98)
2
with m = 1, . . . , N defines the position of the particle. The general Bethe state is given by the
linear combination:

1 ,...,r (x1 , . . . , xr ) =
(99)
Ap1 ,...,pr p1 (x1 ) . . . pr (xr ),
P
where the sum is over all permutations P = {p1 , p2 , . . . , pr } of the integers 1, . . . , r, and
x1 , . . . , xr are the positions of the particles. The coefficients Ap1 ,...,pr are given in terms of the
scattering amplitudes sj,l :

Ap1 ,...,pr = P
(100)
spj ,pi ,
i<j
1
sj,l = sinh (j l 2i ),
(101)
2
where P is the signature of the permutation {p1 , . . . , pr }. In particular, when the roots j sit on
the two lines Im() = /2, we introduce another notation for the Bethe state:
(1 ,...,r+ |1 ,...,r ) 1 +i/2,...,r+ +i/2,1 i/2,...,r i/2 .
(102)
Symmetries By construction, the states 1 ,...,r are antisymmetric under the exchange of
the j . A shift j j + 2i results in a phase factor:
1 ,2 ,...,j +2i,...,r = (1)r 1 ,2 ,...,j ,...,r .
Action of C
(103)
The action of the operator C on the Bethe states is a shift j j + i :
C1 ,...,r = (i)r 1 +i,...,r +i .
(104)
The operator C, acting on the states (102), exchanges the two lines:
C(1 ,...,r+ |1 ,...,r ) = e
i
2
(r+ r )
(1 ,...,r |1 ,...,r+ ) .
(105)
A special class of states (to be discussed in Section 3.5) is defined by the following constraint on
the Bethe integers:
r+ = r = r,
j {1, . . . , r}
(106)
Ij+
= Ij
= Ij .
(107)
These states are eigenvectors of the operator C, with eigenvalue one:

C(1 ,...,r |1 ,...,r ) = (1 ,...,r |1 ,...,r ) .
(108)
501
The other eigenvectors of C are given by the linear combinations:
(
1 ,...,r
+ |1 ,...,r )
C(
1 ,...,r

i
1
= (1 ,...,r+ |1 ,...,r ) e 2 (r+ r ) (1 ,...,r |1 ,...,r+ ) ,
2
) = (1 ,...,r |1 ,...,r ) .
+ |1 ,...,r
(109)
(110)
Action of the shift operator The shift operator eiP has a similar action on the Bethe states:
r

ieip1 (j +i) 1 +i,...,r +i ,
eiP 1 ,...,r =
(111)
j =1
where

sinh 12 ( + i )
exp ip1 () =
.
sinh 12 ( i )
(112)
3.3. Eigenvalues
The eigenvalue of the transfer matrix T (u) associated to the Bethe state 1 ,...,r is:
(1 , . . . , r |u) = (1 , . . . , r |u)(1 + i, . . . , r + i|u),
where
(113)
r

N
sinh 12 (j 2iu i )
sin 2(u )
sinh 12 (j 2iu + i )
j =1

r
1

sinh
(
2iu
+
3i
)
j
2
+ (sin 2u)N
.
1
sinh
2 (j 2iu + i )
j =1
(1 , . . . , r |u) (i/2)
(114)
In the limit u 0, the energy of the state 1 ,...,r is defined as:

1
log
sin 2
(1 , . . . , r |0)
2
u
r

2 sin2 2
.
= 2N cos 2 +
cosh 2j cos 2
E(1 , . . . , r ) =
(115)
(116)
j =1
By construction, the state 1 ,...,r is an eigenvector of the Hamiltonian H, with eigenvalue

E(1 , . . . , r ).
3.4. Reminder: The homogeneous six-vertex model
Consider the homogeneous six-vertex model defined by the R0 (u0 ) matrix given in Eq. (14),
with parameter 0 , on a lattice of width N .
In the anisotropic limit u0 0, the logarithmic derivative of the transfer matrix T0 (u0 ) is
equal to the XXZ Hamiltonian:
502

1 T0
H0 sin 0 T0 (0)
(0)
u0
=
(117)
N

0
cos 0 Em
,
(118)
m=1
0 is a generator of the TemperleyLieb algebra with parameter , like in
where the operator Em
0
Eq. (24):

1 x x
y y
z
z
(119)
.
1 i sin 0 mz m+1
m m+1 + m m+1 cos 0 mz m+1
2
On a lattice with periodic boundary conditions, the system is solvable by Bethe ansatz. The
Bethe equations for r particles are:
0
Em
=
j {1, . . . , r}
r

exp iNk0 (j ) =
exp i0 (j , l ) ,
(120)
l=1
where the one-particle momentum and the scattering amplitude are:

sinh( 2i 0 )

,
exp ik0 () =
sinh( 2i 0 + )

sinh( + i0 )
exp i0 (, ) =
.
sinh( i0 )
The associated eigenvalue of the transfer matrix T0 (u0 ) is:

r

N
sinh(j + iu0 + 2i 0 )
0 (1 , . . . , r |u0 ) = sin(0 u0 )
sinh(j + iu0 2i 0 )
j =1

r

sinh(j + iu0 3i2 0 )
+ (sin u0 )N
.
sinh(j + iu0 2i 0 )
j =1
(121)
(122)
(123)
3.5. Common eigenvalues between the staggered and homogeneous six-vertex models
If the parameters of the staggered and the homogeneous six-vertex models are related by:
0 = 2 ,
(124)
u0 = 2u
(125)
then one has the relations:

exp 2i k()
= exp ik0 () ,
(126)
1 (, ) + 1 (, ) = 0 (, ).
(127)
These relations suggest that the states:

(1 ,...,r |1 ,...,r )
(128)
have properties described by the homogeneous six-vertex model. The states (128) are obtained
as the solution of the Bethe equations (91)(92) when the Bethe integers on the two lines are the
503
same:
r+ = r = r,
(129)
j {1, . . . , r}
Ij+
= Ij
= Ij .
(130)
As a consequence of relations (126)(127), the Bethe equations (91)(92) are equivalent to the
Bethe equation (120) for the homogeneous six-vertex model. Moreover, the eigenvalue of the
transfer matrix T (u) can be written in terms of the eigenvalue (123):

2
1 + i , 1 i , . . . , r + i , r i |u = (1/4)N 0 (1 , . . . , r |u0 ) .
2
2
2
2
(131)
As a consequence, in the anisotropic limit, the energies E, E0 of the Hamiltonians H, H0 are
related by:

= 2E0 (1 , . . . , r ).
E 1 + i , 1 i , . . . , r + i , r i
(132)
2
2
2
2
The total momenta are equal:
ktot = k0,tot .
(133)
4. Finite-size study of the Bethe equations

4.1. Reminder: The XXZ spin-chain Hamiltonian
4.1.1. Definition
The XXZ Hamiltonian (118) on a lattice of width N can be written:
H0 = 1/2
N

y y
z
x
mx m+1
+ m m+1 cos 0 mz m+1
+1 ,
(134)
m=1
0 0
(135)
with periodic boundary conditions:
N +1 = 1 .
(136)
4.1.2. Ground state

The ground state of the Hamiltonian (134) is given by the symmetric half-filled Fermi sea
in the sector with r = N/2 particles. Consider one of the particlehole excitations around the
Fermi momenta kF :
kF k0 /2 kF + k0 /2,
(137)
kF + k0 /2 kF k0 /2.
(138)
These excited states have total momentum k0 . In the thermodynamic limit N with the
ratio r/N fixed, the Bethe equations (120) give a linear integral equation for the density of particles. Solving this equation gives access to the dispersion relation of the excitations (137)(138):
E0 (k0 ) v0 |k0 |,
v0 =
sin 0
.
0
(139)
504
This shows that, in the thermodynamic limit, the spectrum has no finite gap above the ground
state. This is an indication that the equivalent (1 + 1)-dimensional quantum field theory is conformally invariant. The velocity v0 appears in the finite-size determination of the central charge:

v0 c
E0,gr (N ) = Ne0,
(140)
+ o N 1 .
6N
The central charge of the Hamiltonian H0 is c = 1. The energy density per site is:

e0, = cos 0 2 sin 0
2
dx
.
cosh(x)[cosh(20 x) cos 0 ]
(141)
4.1.3. Low-lying excitations

A class of Bethe states are low-lying excitations, with energies:

2v0 x
+ o N 1 ,
(142)
N
where x is called the physical exponent. We describe the primary states, and then their descendants. In the sector with r = N/2 n particles, the lowest-energy state is given by the distribution
of the Bethe integers which is symmetric around zero. In the same sector, denote by {n, m} the
state obtained after m backscatterings from the left Fermi level to the right Fermi level:
E0 (N ) = E0,gr (N ) +
r 1
r 1
+ m, . . . ,
+ m.
2
2
The physical exponents of this class of states are:
I1 , . . . , Ir =
(0)
xn,m
= n2
g
1
+ m2 ,
2
2g
g=
0
.
(143)
(144)
Descendant states are obtained by performing particlehole excitations on the above states. For
example, starting from the ground state distribution and setting Ir to (r + 1)/2, one obtains a
state with physical exponent x = 1.
4.2. Ground state of the Hamiltonian H
In this section, the ground state of the Hamiltonian (74), corresponding to the anisotropic
limit of the staggered six-vertex model, is discussed. The system width N is assumed to be
even. The case of an odd system width N will be discussed in Section 4.4. The ground state
of the Hamiltonian H is given by the symmetric distribution of the Bethe integers in the sector
r+ = r = N/2:
N/2 1 N/2 1
N/2 1
(145)
,
+ 1, . . . ,
,
2
2
2
N/2 1 N/2 1
N/2 1
=
,
+ 1, . . . ,
.
I1 , . . . , IN/2
(146)
2
2
2
Note that this state is of the type described in Section 3.5, and thus it has twice the energy
of the ground state of HXXZ , and the same total momentum. Using this remark, we deduce the
Fermi velocity of particlehole excitations, and the central charge of the ground state.
Consider the double particlehole excitation obtained by performing the transformation (137)
on both Fermi seas. This excitation is of the type (129)(130), and therefore, using Eqs (139),
+
I1+ , . . . , IN/2
=
505
(133) and (132) its momentum and energy are:

k = k0 ,
(147)
E(k) 2v0 |k0 |.
(148)
Thus, the velocity of the particlehole excitations around the ground state of the Hamiltonian H
is:
v = 2v0 .
(149)
Since the ground state energy of the Hamiltonian H is exactly twice the ground state energy of
the Hamiltonian HXXZ , the asymptotic behaviour of the former has the form given by conformal
invariance:

cv
Egr (2N) = 2N e
(150)
+ o N 2 ,
6 2N
where the central charge is:
c=2
(151)
and the energy density per site is:

e = e0, = cos 0 2 sin 0
2
dx
.
cosh(x)[cosh(20 x) cos 0 ]
(152)
4.3. Low-energy spectrum of the Hamiltonian H

4.3.1. Bethe excitations
The excitations considered are combinations of particle and backscattering excitations on the
Bethe integers Ij+ , Ij . Denote (n+ ,n ),(m+ ,m ) the Bethe state defined by the numbers of particles:
r+ = N/2 n+ ,
(153)
r = N/2 n
(154)
and the Bethe integers:

r+ 1
r+ 1
+ m+ , . . . ,
+ m+ ,
(155)
2
2
r 1
r 1
I1 , . . . , Ir =
(156)
+ m , . . . ,
+ m .
2
2
The states described in Section 3.5 are of the type (n,n),(m,m) . Their physical exponents are:
I1+ , . . . , Ir++ =
1
(0)
x(n,n),(m,m) = 2xn,m
= n2 g + m2 ,
g
(157)
where
g = 2 /.
(158)
506
The analytical study [9] of the thermodynamic limit of the Bethe equations suggests the following
general form for the physical exponents:
1
1
x(n+ ,n ),(m+ ,m ) = g(n+ + n )2 + (m+ + m )2
4
4g
1
1
+ K( , N )(n+ n )2 +
(m+ m )2 ,
4
4K( , N)
(159)
where the coupling constant K( , N) tends to zero as N goes to infinity. One of the key questions
is to determine the way this constant actually vanishes.
4.3.2. General structure of the low-energy spectrum
Note that the states with m+ = m are not part of the low-energy spectrum. Thus, in the
following, the discussion will concern only the states with m+ = m = m. For these states, the
total magnetization and the total momentum read:
S = n + + n ,
ktot = m(N n+ n ).
N
(160)
(161)
Eq. (159) determines the structure of the low-energy spectrum. The quantities (n+ + n ) and
m define a sector of given total spin S and total momentum ktot . We call floor states the lowest
, where n, m are
energy states of these sectors. These are the states (n,n),(m,m) and (n+1,n),(m,m)
integers. The physical exponents of the floor states are finite in the limit N :
x(n,n),(m,m) = gn2 +
x(n+1,n),(m,m) =
m2
,
g
g
m2 K( , N)
(2n + 1)2 +
+
.
4
g
4
(162)
(163)
Note that the states (n,n),(m,m) are those described in Section 3.5. Within each sector, higher
energy states are obtained, starting from the floor state, by moving n particles from the line
Im() = /2 to the line Im() = /2 (or the reverse). The physical exponent of the resulting
state differs from the floor exponent by a quantity proportional to the coupling constant K( , N):
x(n+n ,nn ),(m,m) = x(n,n),(m,m) + (n )2 K( , N),
(164)
x(n+n +1,nn ),(m,m) = x(n+1,n),(m,m) + n (n + 1)K( , N ).
(165)
Thus, if the constant K indeed vanishes in the limit N , the gaps between the floor state
and the higher states in the sector should also vanish. The structure of the spectrum is illustrated
in Fig. 14.
Numerical calculations are used to confirm the form (159) of the physical exponents, and to
determine the scaling law for K( , N).
4.3.3. Numerical calculation of the physical exponents
Numerical procedure When the Bethe integers Ij are fixed according to Eqs. (155)(156), the
Bethe equations (91)(92) are a set of non-linear equations for the variables j , j . These equations can be solved numerically, using the multidimensional NewtonRaphson method [13]. The
507
Fig. 14. The first levels of the fundamental sector (n+ + n = 0, m = 0) and the sector defined by (n+ + n = 1, m = 0).
The floor states (including the ground state) are represented by bold lines. The energies are rescaled as physical exponents.
following starting point for the algorithm was found empirically to lead to a good convergence:

tan(Ij+ /N)
=
Arctanh
,
init
j
U
(166)
where the additional parameter is set to U = 10. The exponents are then estimated, relatively to
the ground state, or to the floor state, using:
2v
xa ,
2N
2v
(xa xb ),
Ea Eb
2N
where a, b denote any states of the spectrum.
Ea Egr
(167)
(168)
Results for the floor exponents These are shown in Figs. 1517. These results agree with the
form (159).
Results for the gaps inside a given sector These gaps are given in Eqs. (164)(165), as a function of the integer n and the coupling constant K( , N). The first step is to check the dependence
on n . See Figs. 1821.
Results for the coupling constant Eqs. (164)(165) relate the gaps within a given sector to
the coupling constant K( , N). These are used to determine the scaling law of this quantity.
508
Fig. 15. The floor exponents x(n,n),(0,0) as functions of , for a system width N = 1024. The expected values are
x(n,n),(0,0) = gn2 (Eq. (162)).
Fig. 16. The floor exponent x(1,1),(1,1) as a function of , for a system width N = 512. The expected value is
x(1,1),(1,1) = g + g 1 (Eq. (162)).
Theoretical arguments (see Section 5) predict the following form:

K
A
,
[B + log N ]p
(169)
where the exponent p may take the values p = 1, 2. The exponent p is estimated numerically,
assuming that K behaves as Eq. (169) (see Fig. 22). A crossover is observed between the behav-
509

Fig. 17. The floor exponent x(n+1,n),(0,0)
x(n+1,n),(0,0) x(1,0),(0,0) as a function of , for a system width N = 1024.

The expected value is x(n+1,n),(0,0) = gn(n + 1) (Eq. (163)).
Fig. 18. The exponent x(n,n),(0,0) as a function of n, for parameter = 0.3 . The expected values are
x(n,n),(0,0) = K( , N)n2 (Eq. (164)).
iours:
K( , N)
A( )
[B( ) + log N ]2
K(/2, N)
A
.
B + log N
( < /2),
(170)
(171)
510
Fig. 19. The exponent x(1+n,1n),(0,0) x(1+n,1n),(0,0) x(1,1),(0,0) as a function of n, for parameter = 0.3 . The
expected values are x(1+n,1n),(0,0) = K( , N)n2 (Eq. (164)).
Fig. 20. The exponent x(1+n,1n),(1,1) x(1+n,1n),(1,1) x(1,1),(1,1) as a function of n, for parameter = 0.4 . The
expected values are x(1+n,1n),(1,1) = K( , N)n2 (Eq. (164)).
The factor A( ) in the scaling law (170) is estimated numerically by eliminating the term
B( ) between two system widths. The finite-size estimators are defined as:

A( , N1 , N2 )
log(N1 /N2 )
[K( , N1 )]1/2 [K( , N2 )]1/2
2
.
(172)
511
Fig. 21. The exponent x(1+n,n),(0,0) x(1+n,n),(0,0) x(1,0),(0,0) as a function of n, for parameter = 0.3 . The
expected values are x(1+n,n),(0,0) = K( , N)n(n + 1) (Eq. (165)).
Fig. 22. Estimation of the exponent p in the scaling law (169). The estimator for p is obtained by eliminating the
unknowns A, B from the equations log K( , N) = log A p log(B + log N ), for three system widths. The estimators
converge slowly to p = 2 for < /2, and to p = 1 for = /2.
The numerical results (see Fig. 23) allow us to conjecture the following form for the factor A( ):
A( ) =
5
.
2
(173)
The correction term B( ) depends on the sector considered.

Each sector determines a specific scaling law for K( , N), and these determinations can be
compared. In addition, the coupling constant K( , N) appears in the expression of infinite
512
Fig. 23. Determination of the factor A( ) in the scaling law (170), using a numerical estimation of exponent x(1,1),(0,0) .
Fig. 24. Various determinations of the factor A( ), for system widths N = 256512.
physical exponents:
x(n,n),(1,1) x(n,n),(0,0) =
1
.
K( , N)
(174)
The results are shown in Fig. 24.

4.4. The case N odd
When the system width N is odd, the structure of the spectrum exhibits some differences from
the case N even.
513
Fig. 25. Determination of the factor A( ) in the scaling law (170) by the effective central charge c for N odd.
The general formula for the finite-size corrections is identical to the formula for N even:
E(n+ ,n ),(m+ ,m ) (2N ) = 2N e

v
2v
2+
x(n+ ,n ),(m+ ,m ) + o N 1 , (175)
6 2N
2N
where the exponent x(n+ ,n ),(m+ ,m ) is given by Eq. (159). The subtlety is that Eqs. (153)(154)
imply that n+ , n are half-odd integers.
The lowest-energy states of the spectrum (175) are (1/2,1/2),(0,0)

. Note that these states
belong to the sector S = 0, and are the most closely packed to the imaginary axis in this sector.
The energy of the ground state is:

v
Egr (2N) = 2N e
(176)
c + o N 1 ,
6 2N
where the effective central charge is:
K( , N)
(177)
.
4
The effective central charge is estimated numerically using Eq. (176) and the exact expression (141) of the energy density e (see Fig. 25). The scaling law for the coupling constant
K( , N) is consistent with Eq. (170). It is convenient to define the physical exponents of the
excited states with respect to the ground state:
c = 2 12x(1/2,1/2),(0,0) ,
x(1/2,1/2),(0,0) =
x(n+ ,n ),(m+ ,m ) x(n+ ,n ),(m+ ,m ) x(1/2,1/2),(0,0) .
(178)
The physical exponents of the floor states are:

x(n+1/2,n1/2),(m,m) = gn2 +
x(n1/2,n1/2),(m,m) =
m2
,
g
g
m2 K
(2n 1)2 +
.
4
g
4
(179)
(180)
514
Fig. 26. The floor exponents x(n+1/2,n1/2),(0,0) as functions of , for a lattice width N = 1023. The expected values
are x(n+1/2,n1/2),(0,0) = gn2 (Eq. (179)).

Fig. 27. The floor exponents x(n1/2,n1/2),(0,0)
x(n1/2,n1/2),(0,0) x(1/2,1/2),(0,0) as functions of , for a lattice

= gn(n 1) (Eq. (180)).
width N = 1023. The expected values are x(n1/2,n1/2),(0,0)
In particular, the states (1/2,1/2),(0,0) , (1/2,1/2),(0,0) have an energy which is twice that of the
degenerate ground state of the XXZ spin-chain (134) with an odd lattice width [14]. These two
states appear as excited states of the Hamiltonian H.
The gaps inside a given sector are:
x(n+n +1/2,nn 1/2),(m,m) = x(n+1/2,n1/2),(m,m) + Kn (n + 1),
2
x(n+n 1/2,nn 1/2),(m,m) = x(n1/2,n1/2),(m,m) + K(n ) .

Expressions (179)(182) are checked on Figs. 2629.
(181)
(182)
515
Fig. 28. The exponent x(1/2+n,1/2n),(0,0) x(1/2+n,1/2n),(0,0) x(1/2,1/2),(0,0) as a function of n, for parameter = 0.3 . The expected values are x(1/2+n,1/2n),(0,0) = K( , N)n(n + 1) (Eq. (181)).
Fig. 29. The exponent x(1/2+n,1/2n),(0,0) x(1/2+n,1/2n),(0,0) x(1/2,1/2),(0,0) as a function of n, for parameter
= 0.3 .The expected values are x(1/2+n,1/2n),(0,0) = K( , N)n2 (Eq. (182)).
5. Interpretation and relation to non-linear sigma models

As discussed in Section 2, the limit /2 of the staggered six-vertex model coincides with
a particular point of the OSP(2|2) loop model of [7]. This is a good starting point to understand
the emergence of a continuous spectrum of critical exponents.
516
Fig. 30. Interaction vertices in the OSP(m|2n) symmetric sigma model on the square lattice.
5.1. A reminder on intersecting loop models and Goldstone phases

It turns out that the MerminWagner theorem forbidding the spontaneous breaking of a continuous symmetry in two dimensions does not hold for supergroups (because of the lack of
unitarity), and that models with orthosymplectic OSP(m|2n) symmetry do exhibit a low temperature phase with spontaneous broken symmetry provided m 2n < 2. More precisely, consider the
non-linear sigma model with target space the supersphere S m,2n = OSP(m|2n)/OSP(m 1|2n),
a supersymmetric extension of the usual O(N) sigma model. Use as coordinates a real scalar
field:
(1 , . . . , m , 1 , . . . , 2n )
(183)
and the invariant bilinear form

=
a a +
J ,
(184)
where J is the symplectic form which we take consisting of diagonal blocks:

supersphere is defined by the constraint:
0 1
.
1 0
The unit
= 1.
The action of the sigma model (conventions are that the Boltzmann weight is

1
d 2 x .
S=
2g
The perturbative function depends only on m 2n to all orders:

(g ) = (m 2n 2)g2 + O g3 .
(185)
eS )
reads:
(186)
(187)
The model for g positive thus flows to strong coupling for m 2n > 2. Like in the ordinary sigma models case, the symmetry is restored at large length scales, and the field theory
is massive. For m 2n < 2 meanwhile, the model flows to weak coupling, and the symmetry is
spontaneously broken. One expects this scenario to work for g small enough, and the Goldstone
phase to be separated from a non-perturbative strong coupling phase by a critical point.
It is easy to suggest lattice models whose long distance physics is described by the supersphere
sigma models. It was shown in [7] that simply taking a square lattice with the fundamental
representation of the OSP(m|2n) on each link and Heisenberg coupling at every vertex would
suffice. More generally, it is convenient to represent the link states by lines carrying a label
a = 1, . . . , m or = 1, . . . , 2n and express the interactions in terms of the three invariant tensors
of the algebra, corresponding to the three diagrams in Fig. 30.
A trivial rescaling of the partition function and isotropy of the model allows one to set the first
two weights equal to 1. The remaining weight is a free bare parameter. It was found in [7] that for
any w > 0 and m 2n < 2 the lattice model is indeed described at large distance by the UV limit
of the sigma model, i.e., a set of m 1 free uncompactified bosons and n pairs of symplectic
517
fermions, with a central charge c = m 2n 1. The value w = 0 is critical, and was argued in
[7,15] to correspond to the critical point of the sigma model. Note that for m 2n fixed, there
exists a particular value of the crossing weight w where the model is integrable, and coincides
with the one in [12].
Of course, the geometrical representation of the invariant tensors allows for a full geometrical
formulation of the model, that can then be extended to values of m, n not integer. The model
so obtained is made of loops covering every link of the lattice, and possibly self-intersecting
once on the vertices, with a fugacity m 2n, and a weight w per crossing. The continuum limit
was found to be a naive extrapolation from the results at m, n integer. Note thatthis model
differs from the usual formulation of the Q-state critical Potts model with m 2n Q only in
that crossings are allowed. This however changes the universality class deeply. For instance, all
the L-leg operators
which have non-trivial, Q-dependent scaling dimensions in the Potts model
for 2 Q 2, now have logarithmic correlators with vanishing effective dimension. This
is because the symmetry being spontaneously broken the fundamental field has non-vanishing
expectation value, and thus the fields a do exist in the conformal field theory, unlike in the case
of a compact boson.
[Paths that can self intersect at a vertex but not at a bond are called trails in the literature, and
bond self avoiding models by contrast with site self avoiding. The case m 2n = 0 we consider below would correspond to fully packed trails. People have considered ordinary (dilute)
trails which are in the same universality class as dilute SAW and trails at the theta point. Those
have been shown to be in the same universality class as growing self avoiding trails, themselves
a particular case of the trajectory of a particle moving on a lattice with random distribution of
scattering rotators. See [16,17]. In the context of such trajectories, it has been remarked already
that when one dilutes the set of random rotators from maximum concentration CL = CR = 1/2
(i.e., add intersections), the exponents change.]
5.2. Coupling constant and physical exponents of the OSP(2|2) model
We now specialize to the case of interest here, the OSP(2|2) model. The UV limit is easy to
understand. We can parametrize the supersphere
(1 )2 + (2 )2 + 21 2 = 1
(188)
by setting
1 = cos (1 1 2 ),
2 = sin (1 1 2 ),
+ 2.
The action then reads

1
S=
d 2 x ( )2 (1 21 2 ) + 2 1 2 41 2 1 2 .
2g
(189)
(190)
The coupling g > 0 flows to zero at large distances. On the other hand, we can absorb it by
rescaling all fields so the action reads

1
S=
(191)
d 2 x ( )2 (1 2g 1 2 ) + 2 1 2 + 4g 1 2 1 2 ,
2
where now has a different radius, + 2g . We see that as g 0 all interaction terms
disappear and we get a free boson together with a pair of free symplectic fermions 1,2 .
518
Moreover the radius of compactification goes to infinity in that limit, so the boson appears as
non-compact.
This holds in the true large distance limit. At intermediate scales, we can use the RG equation
for the coupling [18]
1
m 2n 2 2
dg
=
g = g2 .
d log l
2
(192)
Writing more generally

dg
= g2
d log l
(193)
we see that g approaches its vanishing large distance value as

1
1
= 0 + log(l/ l0 ) log(l/ l0 ).
g
g
(194)
Here, l is a characteristic dimensionless scale ratio, roughly of the order of the ratio of the scale
at which one is observing the physics to the lattice cut-off. On the cylinder, l can be identified
with the width in lattice units, l = N .
In the limit of large l, we can estimate more precisely the contribution to the spectrum coming
from
. Recall that for a free bosonic theory where the action is normalized as S =
the boson
1
2 and the field compactified as X X + 2R, the spectrum of dimensions [19] is
X)
(
8
m2 R 2
e2
.
+ = 2 +
4
R
(195)
Matching the normalization gives R 2 = 4/g in our case, and thus we expect the scaled gaps
coming from the bosonic degrees of freedom to read, at large distances:
+ =
e2
+ m2 log(l/ l0 ).
4 log(l/ l0 )
(196)
In the limit l the dimensions become degenerate and the spectrum can be considered as a
continuum starting above + = 0. To emphasize the latter point, consider the contribution to
the partition function coming from the degrees of freedom. The system is defined on a torus of
periods l and l , with l / l = . Denote q = exp(2i ).
1 (e/R+mR/2)2 /2 (e/RmR/2)2 /2
Z =
q
q
e,m

R
R 2 |m m |2
1
=
exp
2 Im
2 Im

m,m
R
1

,
R
2 Im

n
1/24 P (q). Observe now that one can write:
where = q 1/24
n=1 (1 q ) = q
1
=4
Im

0
(197)
q s q s
ds
(198)
519
which can be interpreted as an integral over a continuum of critical exponents = = s 2 . In

the partition function (197), R plays the role of the density of levels, and is proportional to the
(diverging) size of the target space.
Going back to the specific case of the OSP(2|2)/OSP(1|2) model, we set = 1/ . We get
the radius R 2 = 4 log(l/ l0 ), and the contribution of the free boson to the spectrum is:
+ =
(e2 )2
+ (m2 )2 log(l/ l0 )
4 log(l/ l0 )
(199)
with e2 , m2 arbitrary integers. Suppose now we consider the sigma model with a combination of
periodic and antiperiodic boundary conditions for the symplectic fermions. These get untwisted,
and add to Z the discrete spectrum of a free boson at the free fermion point, which is given by
Eq. (195) with a radius R 2 = 1. The quantum numbers associated to the fermions will be denoted
e1 , m1 . Thus we expect:
(m1 )2
(e2 )2
+ = (e1 )2 +
+
+ (m2 )2 log(l/ l0 ).
4
4 log(l/ l0 )
(200)
We can now compare with formula (159). It is not entirely clear how constrained the quantum
numbers might be in the lattice realization we are considering. But observe that finite scaled gaps
occur for m+ = m = m and converge to
x=
m2 1
1
+ g(n+ + n )2 + K( , N )(n+ n )2 .
g
4
4
(201)
When tends to /2, Eqs. (158) and (171) give the coupling constants:
g = 1,
K(/2, N)
A
.
log N
(202)
The numerical results are compatible with A = 1. This suggests the identifications:
1
e1 = (m+ + m ) = m,
2
m1 = n+ + n ,
e2 = n + n ,
1
m2 = (m+ m ) = 0.
2
Note that the quantum numbers are (weakly) correlated: the integers m1 , e2 are such that (m1 +
e2 ) is even.
5.3. Interpretation of the staggered six-vertex model for < /2
In preparation for the subsequent discussion, we will denote the effective (square) radius of
the non-compact boson by (R2 )2 (equal to 4 log(l/ l0 ) for = /2) and the radius of the compact
boson by (R1 )2 (equal to 1 for = /2). Away from = /2, it would be reasonable to try and
interpret our results in terms of a deformation of the supersphere sigma model. Several scenarios
are a priori possible. From the numerical results, the compact direction clearly has a modified
radius which becomes:
2
.
(R1 )2 = g =
(203)
520
As for the non-compact direction, a first possibility would be to consider a constant ( ) in the
foregoing discussion. Although it provides reasonable results, much better fits are obtained with
a radius going like log(l/ l0 ) for = /2; specifically, we find:
(R2 )2 =
2
2
log(l/ l0 ) .
5
(204)
Note that this becomes ill-defined as tends to /2, where there is crossover to a behaviour
linear in log(l/ l0 ).
The most naive interpretation of the corresponding target space would be a torus with one
period diverging like log(l/ l0 ) in the scaling limit. This is reminiscent of results on the sausage
model [20] which is a deformation of the usual sphere sigma model. There however, the theory instead of being massless in the IR is massless in the UV, while the target space in that
limit is asymptotically a cigar. The dependence of the cigar dimensions on the RG scale and the
anisotropy parameter are however reminiscent of ours; in particular the long dimension goes
as the logarithm of the RG coordinate in both cases.
Another, more suggestive interpretation, can be obtained if we consider the partition function
2
of our model on the torus. Set = /t so (R2 )2 = t2
5 [log(l/ l0 )] . Using the continuum representation (197) and reabsorbing the t 2 prefactor that comes from the dependence of R2 upon t
into the continuously varying exponents we obtain the partition function:

(mte)2
(m+te)2
s2
s2
log(l/ l0 )
t2 +
4t
t2 +
4t
Z
ds
q
q
,
()
2 e,m=
(205)
where the proportionality

constant is a presumably non-universal quantity, independent of t ,
24
equal approximately to 1/ 10. Observe now that 2(t+1)
t2 4(t2) = 2 so the conformal weights
can be written as well, with respect to c =
2(t+1)
t2 ,
as:
s 2 + 14
(m te)2
(206)
+
.
t 2
4t
This coincides with the contribution of the continuous representations to the spectrum of the
Euclidian black hole CFT SL(2, R)/U (1) coset model [21].
It might be useful here to give some quick reminders for the spectrum of this model. The
central charge for level k is:
h=
2(k + 1)
(207)
.
k2
Normalizable operators come in two kinds. There is the ones (delta functions normalizable)
associated with the principal continuous series j = 12 + is with conformal weight:
cBH =
h=
j (j + 1) (n kw)2 s 2 + 14
(n kw)2
+
=
+
k2
4k
k2
4k
(208)
1
and the ones coming from the discrete series. They have j [ 1k
2 , 2 ] together with rules relating allowed values to w, n. The net result is that these fields all have dimensions larger than the
bottom of (208). Note that the identity field h = 0 is not among the normalizable states, which is
consistent with the fact that we do not observe (after the obvious identification k t ) cBH in the
24
lattice model but cBH 4(k2)
= 2.
521
Of course the partition function of the Euclidian black hole theory should naively be infinite
since it involves infinite dimensional representations of SL(2). The introduction of a Liouville
wall at finite distance in the target space [22] gives a density of states (s) log to leading
order, which agrees with our results provided we identify l/ l0 , and thus the size of the
system (it would certainly be interesting to investigate the subleading behaviour, which depends
on s, in the lattice data).
Finally, we note that in [22], the sum (4.17) implies some combinatorial constraints on the
descendents, which we have not studied in the lattice model.
Note that further twisting and reduction of the vertex model gives rise to parafermionic theories, which are themselves cosets SU(2)/U (1). It is not clear how this might be related to the
identification of the untwisted vertex model with SL(2, R)/U (1).
6. Geometrical interpretation of the critical exponents
For the OSP(m|2n) models in their Goldstone phases, the continuous spectrum of critical exponents has its origin (like in the case of a pure non-compact boson) in the existence of infinitely
many operators with vanishing scaling dimension (the powers of in the case of the non-compact
boson), which in itself is a consequence of the spontaneously broken symmetry [7]. An obvious
question is whether a similar interpretation exists for our model.
Some insight can be gained by considering the limit /2 which is related to the OSP(2|2)
model. In the latter, exponents of the order parameter are related with geometrical correlations
of the degrees of freedom carrying lines, and therefore it is tempting to ask whether this might
hold away from the limit point as well. Before discussing this, it is important to notice that
the staggered six-vertex model at = /2 (and the equivalent 38-vertex model) with periodic
boundary conditions correspond, strictly speaking, to the model of [7,10] where the OSP symmetry is broken by the boundary. This is the reason why the spectrum of conformal weights
contains a compact boson with (R1 )2 = 1, and thus non-trivial, finite exponents. Within the
geometrical interpretation to be discussed below, this will correspond to the existence of some
correlators having non-trivial weights, while others do behave as in [7]. Let us now be more
specific. As shown in Section 2.3, summing on 2 2 vertex blocks and choosing the right
basis, the staggered six-vertex model is equivalent to the 38-vertex model. Any lattice configuration within this model is completely covered by polygons of three different types: oriented
lines, bare thin lines and bare thick lines. The particular point = /2, as well as our discussion on Goldstone phases, suggests to consider geometrical correlations associated with these
lines.
6.1. Correlation functions with no thin or thick line
Consider first the watermelon correlation function (as illustrated on Fig. 31) consisting of a
forced even number 2r of positively-oriented lines (that is, we force 2r such lines to propagate
through the system, on top of fluctuating numbers of positively and negatively oriented lies in
equal number and bare lines). The critical exponent of this correlation function is given by the
lowest-energy state in the sector with fixed total spin S = 2r and C = 1. According to Eq. (159),
this state is (r,r),(0,0) . According to Eq. (105), this is an eigenstate of the operator C, with
eigenvalue one.
522
Fig. 31. Some watermelon correlation functions. (a) No bare line, (b)(c) one bare line, (d) one thin line and one thick
line.
6.2. Correlation functions with one thin or thick line

Consider the correlation function consisting of a forced odd number 2r 1 of positivelyoriented lines, along with one bare (thin or thick) line. The critical exponent of the correlation
function with a thin (respectively thick) line is given by the lowest-energy state in the sector with
fixed total spin S = 2r 1 and C = 1 (respectively C = 1). Both states (r,r1),(0,0)

have the
physical exponent x(r,r1),(0,0) . Thus, the correlation functions with one thin or thick line have
the same exponent x(r,r1),(0,0) .
6.3. Correlation functions with one thin line and one thick line
Consider the correlation function consisting of a forced even number 2r of positively-oriented
lines, along with one thin line and one thick line. The critical exponent of this correlation function
is given by the lowest-energy state in the sector with fixed total spin S = 2r and C = 1. This
state is (r+1,r1),(0,0)
, with physical exponent x(r+1,r1),(0,0) .
In the particular case r = 0, this exponent becomes:
x(1,1),(0,0) = K( , N ).
(209)
This exponent vanishes in the thermodynamic limit. So the two-leg correlation function consisting of one thin line and one thick line belongs to the continuous subspectrum associated to the
ground state.
6.4. Interpretation
The physical interpretation we suggest for the continuum limit of the 38-vertex model is thus
the following. The arrow degrees of freedom can be treated like the usual domain boundaries
for an RSOS model which renormalizes, at large distances, to a compactified free bosonic field
with radius R1 . A correlation function involving a certain number S of positively oriented lines
corresponds for this bosonic field to a magnetic or vortex operator, whose physical exponent is
given by x = gS 2 /4.
Meanwhile, correlators involving in addition thin and thick lines as well have, in the thermodynamic limit, exponents entirely determined from the contribution of the arrow degrees of
523
freedom.1 Thin and thick lines correspond thus to operators with vanishing exponents and (presumably) logarithmic correlators, and behave similarly to the crossing lines in the models of [7].
7. Conclusion, open problems
First and foremost, we believe this work puts on firm ground the existence of a continuous
spectrum of critical exponents in a model with a finite number of lattice degrees of freedom per
site (link), justifying fully the conclusions of [7,8].
It is truly remarkable that a proper staggering of the simple six-vertex model could give rise
to such interesting behaviour: obvious directions for future work are plenty, and include an
analytic derivation of the coupling constant K, a better understanding of the relationship with
SL(2, R)/U (1), and of the effect of the various twist and truncations necessary to produce the
Potts and RSOS versions of this model.
It should also be possible to generalize the problem to some higher spin version. Some comments on Bethe equations are in order here. Recall that the usual source term for these equations
reads, in the case of spin 1/2:

sinh 12 ( i ) N
(210)
.
sinh 12 ( + i )
One might think of changing it through real heterogeneities , leading to equations which have
been used a great deal in the study of massive deformations [23]:

sinh 12 ( i ) N/2 sinh 12 ( + i ) N/2
.
(211)
sinh 12 ( + i )
sinh 12 ( + + i )
One could also think of changing it through imaginary heterogeneities. The simplest case would
correspond to adding string heterogeneities, i.e., for the simplest case of the two string, formally = i . Clearly however half the modified source terms cancel out, leaving:

sinh 12 ( 2i ) N
(212)
sinh 12 ( + 2i )
which is nothing but the source term for a spin one chain. In general, heterogeneities of the type
p-string will lead to the equations for the spin p/2 chain. The other natural possibility would
be to add heterogeneities of the anti-string type, i.e., = i . Up to a shift + i , this
produces however the initial equations, and is well known to simply change the Hamiltonian
from ferromagnetic to antiferromagnetic. The possibility we encountered in this paper consists
in adding heterogeneities right in the middle of the physical strip, at = i/2. Note that the
higher spin generalization is obvious by changing i to 2si everywhere.
1 A subtle remark is in order here. Recall the conserved quantities of the 38-vertex model: the total spin and the value
of operator C. In geometrical terms, this means that the total arrow flow is conserved, whereas only the parity of the
number of thick lines is conserved. To define a correlation function with more than one thick lines within the 38-vertex
partition function, it is necessary to define the transfer matrix on a non-local Hilbert space (so that the transfer matrix
(8)
keeps track of the connectivity of the thick lines). In this framework, three interpretations are possible for the vertex a1 .
The behaviour of the correlation functions is likely to depend on the splitting of this vertex into the three possible
connectivities, though we believe that the behaviour is universal, in agreement with this conclusion. We did not enter into
these technicalities here, and simply described the correlation functions which are not affected by this problem.
524
It is also fascinating that the model should interpolate between OSP(2/2) and SO(4) symmetries when goes from /2 to zero. This suggests the existence of a possible quantum group
symmetry all along the line, which we have unfortunately not yet been able to identify.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
R.J. Baxter, Proc. R. Soc. London A 383 (1982) 43.

V. Schomerus, Phys. Rep. 431 (2006) 29.
J. Maldacena, H. Ooguri, J. Math. Phys. 42 (2001) 2929.
M.R. Zirnbauer, Conformal field theory of the integer quantum Hall plateau transition, hep-th/9905054.
B. Duplantier, Phys. Rep. 184 (1989) 229.
A.V. Belitsky, S.E. Derchakov, G.P. Korchemsky, A.N. Manashov, Baxter Q-operator for graded SL(2|1) spin chain,
hep-th/0610332.
J.L. Jacobsen, N. Read, H. Saleur, Phys. Rev. Lett. 90 (2003) 090601.
F. Essler, H. Frahm, H. Saleur, Nucl. Phys. B 712 (2005) 513.
J.L. Jacobsen, H. Saleur, Nucl. Phys. B 743 (2006) 207.
J.L. Jacobsen, H. Saleur, Nucl. Phys. B 716 (2005) 439.
R.J. Baxter, Stud. Appl. Math. 50 (1971) 51.
M.J. Martins, B. Nienhuis, R. Reitman, Phys. Rev. Lett. 81 (1998) 504.
W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes in C, second ed., Cambridge Univ.
Press, 1992.
F.C. Alcaraz, M.N. Barber, M.T. Batchelor, Ann. Phys. 182 (1988) 280.
N. Read, H. Saleur, Nucl. Phys. B 613 (2001) 409.
A.I. Owczarek, T. Prellberg, J. Stat. Phys. 79 (1995) 951.
M.S. Cao, E.G.D. Cohen, J. Stat. Phys. 87 (1997) 147.
E. Abdalla, M.C.B. Abdalla, K.D. Rothe, Non-Perturbative Methods in Two-Dimensional Quantum Field Theory,
World Scientific, 1991.
P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory, Springer, New York, 1997.
V. Fateev, E. Onofri, A.B. Zamolodchikov, Nucl. Phys. B 406 (1993) 521.
R. Dijkgraaf, H. Verlinde, E. Verlinde, Nucl. Phys. B 371 (1992) 269.
A. Hanany, N. Prezas, J. Troost, JHEP 0204 (2002) 014.
See for instance: C. Destri, H. de Vega, Nucl. Phys. B 438 (1995) 413;
N.Yu. Reshetikhin, H. Saleur, Nucl. Phys. B 419 (1994) 507.
Analytical approximation schemes for solving exact

renormalization group equations in the local potential
approximation
C. Bervillier , B. Boisseau, H. Giacomini
Laboratoire de Mathmatiques et Physique Thorique, UMR 6083 (CNRS), Fdration Denis Poisson,
Universit Franois Rabelais, Parc de Grandmont, 37200 Tours, France
Abstract
The relation between the WilsonPolchinski and the Litim optimized ERGEs in the local potential
approximation is studied with high accuracy using two different analytical approaches based on a field
expansion: a recently proposed genuine analytical approximation scheme to two-point boundary value
problems of ordinary differential equations, and a new one based on approximating the solution by generalized hypergeometric functions. A comparison with the numerical results obtained with the shooting
method is made. A similar accuracy is reached in each case. Both two methods appear to be more efficient than the usual field expansions frequently used in the current studies of ERGEs (in particular for the
WilsonPolchinski case in the study of which they fail).
PACS: 02.30.Hq; 02.30.Mv; 02.60.Lj; 05.10.Cc; 11.10.Gh; 64.60.Fr
Keywords: Exact renormalization group; Derivative expansion; Critical exponents; Two-point boundary value problem;
Generalized hypergeometric functions
E-mail addresses: claude.bervillier@lmpt.univ-tours.fr (C. Bervillier), bruno.boisseau@lmpt.univ-tours.fr

(B. Boisseau), hector.giacomini@lmpt.univ-tours.fr (H. Giacomini).
doi:10.1016/j.nuclphysb.2007.07.005
526
C. Bervillier et al. / Nuclear Physics B 789 [FS] (2008) 525551
1. Introduction
The non-decoupling of the relevant scales on a wide and continuous range of magnitudes
in many areas of physics has led to the invention (discovery) of the renormalization group
(RG) [1]. Whereas they have been discovered in the framework of the perturbative (quantum
field) theory, the RG techniques tackle a nonperturbative physical phenomenon [2]. Nonperturbative approaches are difficult to implement and to control, and during a long time one has
essentially carried on perturbative RG techniques (see, e.g., [3]). Nowadays, the huge growth of
the computing capacity has greatly modified this behaviour pattern and, already since the beginning of the ninetys, one has considered [4] with a greater acuteness the exact RG equations
(ERGEs) originally introduced by Wilson [5], Wegner and Houghton [6] in the seventys and
slightly reformulated by Polchinski [7] in the eightys (for some reviews on the ERGEs see [8]).
Initially, the ERGEs are integro-differential equations for the running action S[, t] [assuming
that (x) generically stands for some field with as many indices as necessary and t = ln(/0 )
the logarithm of a running momentum scale ]. They have been extended to the running (average) effective action [, t] [4,9]. Such general equations cannot be studied without the recourse
to approximations or truncations. One of the most promising approximations is a systematic expansion in powers of the derivative of the field (derivative expansion) [10] which yields a set
of coupled nonlinear partial differential equations the number of which grows quickly with the
order of the expansion. In the simplest cases (e.g., for the scalar field), the determination of
fixed points (and of their stability) amounts to study ordinary differential equations (ODEs) with
a two-point boundary value problem that may be carried out numerically via a shooting (or a
relaxation) method.
A pure numerical study is in general not easy to implement and to control. For example, in
the shooting method, the discovery of the right adjustment of the parameters at the boundaries
requires a good knowledge a priori of their orders of magnitude (initial guesses). It is thus interesting to develop concurrently some substitute analytical methods. A popular substitute to the
ODEs of the derivative expansion is provided by an additional expansion in powers of the field
which yields a set of coupled algebraic equations which may be solved analytically, at least with
the help of a symbolic computation software. Various field expansions have been implemented
with more or less success [1114]. Unfortunately, the methods proposed up to now, if they are
easy to implement, do not work in all cases and especially in the most famous and simplest case
of the WilsonPolchinski ERGE [5,7] (equation for the running action S[, t] with a smooth
cutoff).
The object of this paper is to present two new substitute analytical methods for studying
ODEs which, at least in the local potential approximation of the derivative expansion (LPA),
works for the WilsonPolchinski ERGE. One of the methods, recently proposed in [15], is a
genuine analytical approximation scheme to two-point boundary value problems of ODEs. The
other method is new. It is based on approximations of the solution looked for by generalized hypergeometric functions. It has a certain similarity with another new and interesting method based
on the representation of the solution by Pad approximants just proposed in [16] by P. Amore
and F.M. Fernandez independantly from the present work. We illustrate the effectiveness of the
two methods with the explicit consideration of two ERGEs in the local potential approximation:
the WilsonPolchinski equation and the Litim optimized RG equation [17] for the running effective action (named the Litim equation in the following). Following a conjecture first stated in
[18,19], the equivalence of these two equations (in the LPA) has been proven by Morris [20] and
recently been numerically illustrated [21] with an unprecedented accuracy for the scalar field in
527
three dimensions (d = 3). This particular situation provides us with the opportunity of testing
efficiently the various methods of study at hand.
The following of the paper is divided in five sections. In Section 2, we briefly present the
direct numerical integration of the ODEs for the scalar model using the shooting method: determinations of the fixed point and the critical exponents for both the WilsonPolchinski and Litim
equations in the LPA (distinguishing between the even and odd symmetries). A brief presentation of the currently used field expansion is given in Section 3. In Section 4, we analyze several
aspects of the method of [15] applying it to the study of the two equations. We calculate this way
the fixed point locations with high precision and compare the results with the estimates obtained
in Section 2. We show how the leading and the subleading critical exponents may be estimated
using this recent method. In Section 5 we present a new approximate analytical method for ODEs
which is based on the definition of the generalized hypergeometric functions. We show that it is
well adapted to treat the WilsonPolchinski case whereas the Litim case is less easily treated.
We relate these effects to the convergence properties of the series in powers of the field. Finally
we summarize this work and conclude in Section 6.
2. Two-point boundary value problem in the LPA
In this section we briefly present the two-point boundary value problem to be solved in the
LPA of the ERGE. The WilsonPolchinski equation is first chosen as a paradigm in Section 2.1.
The principal numerical results obtained from the numerical integration of the ODE using the
shooting method are given. In Section 2.2, the Litim equation is also studied.
2.1. WilsonPolchinskis flow equation for the scalar-field
The original WilsonPolchinski ERGE in the LPA expresses the evolution of the potential
U (, t) as varying the logarithm of the momentum scale of reference t = ln(/0 ) (with
R). In three dimensions, it reads:
1
U = U (U )2 U + 3U,
2
in which U U (, t)/t , U U (, t)/, U 2 U (, t)/ 2 .
(1)
2.1.1. Fixed point equation

The fixed point equation corresponds to U = 0. It is a second order ODE for the function
U ():
1
U (U )2 U + 3U = 0,
2
(2)
the solution of which (denoted U () below) depends on two integration constants which are
fixed by two conditions. The first one comes from a property of symmetry assumed to be1
U () = U () which provides the following condition at the origin for U ():
U (0) = 0.
1 The other possibility U () = U () gives only singular solutions at finite .
(3)
528
The second condition is the requirement that the solution we are interested in must be nonsingular in the entire range [0, [. Actually, the general solution of (2) involves a moving
singularity [22] of the form:
Using = ln |0 |,
(4)
depending on the arbitrary constant 0 . Pushing 0 to infinity allows to get a non-singular

potential since, in addition to the two trivial fixed points U 0 (Gaussian fixed point) and
2
U 13 + 2 (high temperature fixed point), Eq. (2) admits a non-singular solution which, for
, has the form:
Uasy () =

2
18b2 2/5 1
108b3
+ O 4/5 ,
+ b 6/5 +
+
2/5
2
25
3 625
(5)
in which b is the only remaining arbitrary integration constant. The non-trivial (WilsonFisher
[23]) fixed point solution which we are interested in must interpolate between Eqs. (3) and (5).
Imposing these conditions fixes uniquely the value b of b which corresponds to the fixed point
solution we are looking for.
We have determined b by using the shooting method [24]: starting from a value a supposed to be large where the condition (5) is imposed (with a guess, or trying, value of b b ),
we integrate the differential equation (2) toward the origin where the condition (3) is checked
(shooting to the origin), we adjust the value of b to b so as the latter condition is satisfied with
a required accuracy. A study of the stability of the estimate of b so obtained on varying the
value a provides some information on the accuracy of the calculation.
Rather than (5), it is more usual to characterize the fixed point solution from its small field
behaviour:

3k
k(1 + 3k) 4 k(1 + 3k)(1 + 24k) 6
U () = k 2 +
(6)

+ O 8 ,
2
4
120
and to provide the value of either of the two (related) quantities:
k = U (0),
r =U

(7)
(0) = 3k .
(8)
In the shooting-to-origin method, the determination of r (or k ) is a byproduct of the adjustment of b .

The adjustment of b may be bypassed by shooting from the origin toward a , then r is
adjusted in such a way as to reach the largest possible value of a . In that case b is a byproduct
of the adjustment.
Because the boundary condition at a is under control, the shooting-to-origin method provides
a better determination of r than the shooting-from-origin method. However, this latter method
is more flexible and may easily yield a rough estimate on r which can be used as a guess in a
more demanding management of the method. Notice that, due to the increase of the number of
adjustable parameters, this way of determining a guess is no longer possible in a study involving
several coupled EDOs. Consequently, the development of other methods as, for example, those
two presented below is useful to this purpose (see also [16]).
Table 1 displays the determinations of r and b for three values of a . One may observe
that a high accuracy on r is required to reach a yet small value of a whereas b is only poorly
determined. Obviously, considering higher values of a and/or higher order terms in Eq. (5)
529
Table 1
The fixed point parameter r is already well determined for rather small
values of a whereas b [fixed point value of b in (5)] still is not
r
0.2285982024370220
0.2285982024370219
0.2285982024370219
2.2963
2.3116
2.3162
10
20
40
allows to better determine b , one more term in (5) and a = 1000 yields:
b = 2.31829,
(9)
but the estimate of

is not improved compared to the values given in Table 1 (the machineprecision was already reached). We finally extract from Table 1 our best estimate of r (or k ) as
obtained from the study of the fixed point equation (2) alone:
r = 0.228598202437022 1015 ,
(10)
k = 0.0761994008123407 1016 .
(11)
U ()
the knowledge of which

Individually, these values do not define the potential function
requires the numerical integration explicitly performed in the shooting method.
2.1.2. Eigenvalue equation
The critical exponents are obtained by linearizing the flow equation (1) near the fixed point
solution U (). If one inserts:
U (, t) = U () + et g(),
into the flow equation and keeps the linear terms in , one obtains the eigenvalue equation:

g + (3 )g = 0.
(12)
2
Again it is a second order ODE the solutions of which are characterized by two integration
constants.
Since U () is an even function of , Eq. (12) is invariant under a parity change. Then one
of the integration constants is fixed by looking for either an even or an odd eigenfunction g()
which implies either g (0) = 0 (even) or g(0) = 0 (odd). The second integration constant is fixed
at will due to the arbitrariness of the normalization of an eigenfunction. Thus, assuming either
g(0) = 1 (even) or g (0) = 1 (odd), the solutions of (12) depend only on and on the fixed point
parameter k . For example, these solutions have the following expansions about the origin = 0:

( 3) 2
( 2 12k ) 2
1+
+ O 6 ,
geven () = 1 +
2
12

(2 5 12k ) 3
godd () = +
+ O 5 .
12
When the fixed point solution U is known, the values of [the only remaining unknown parameter in (12)] are determined by looking for the solutions which interpolate between either
g (0) = 0 (even) or g(0) = 0 (odd) and the regular solution of (12) which, for , is:
g 2g U
530
Table 2
Values of the critical exponent determined together with b (and thus r ) whereas a is
varied. Compared to Table 1, a better determination of b is obtained [see the best value
of b given by Eq. (9)]
0.64956177388011
0.64956177388080
0.64956177388065
0.64956177388065
2.318145
2.318257
2.318280
2.318285
12
22
32
40

36b 2 (2 3)
12b
gasy () = S0
1 + (3 )
25 4/5
625 8/5

2(2 1)
,
+
+ O 12/5
125 2
2(3)
5
(13)
in which b is given by (9). The value of S0 is related to the choice of the normalization of
the eigenfunction at the origin, it is a byproduct of the adjustment in a shooting-from-origin
procedure.
In the even case, it is known that the first non-trivial positive eigenvalue 1 (there is also the
trivial value 0 = d = 3), is related to the critical exponent which characterizes the Ising-like
critical scaling of the correlation length . One has = 1/1 and the first negative eigenvalue,
2 , is minus the Ising-like first correction-to-scaling exponent 1 (1 = 2 ) and so on.
In the odd case, the two first (positive) eigenvalues are trivial in the LPA. One has:
d +2
1 =
(14)
,
2
d 2+
,
2 =
(15)
2
in which is the critical exponent which governs the large distance behaviour of the correlation functions right at the critical point, it vanishes in the LPA. With the dimension d = 3 and
the approximation (LPA) presently considered, (14) and (15) reduce to 1 = 2.5 and 2 = 0.5.
Consequently the first non-trivial eigenvalue is negative and defines the subcritical exponent
5 = 1 = 3 sometimes considered to characterize the deviation of the critical behaviour of
fluids from the pure Ising-like critical behaviour.
To determine the eigenvalues we use again the shooting-to-origin method with the two equations (2), (12). However, in addition to , we leave also b adjustable instead of fixing it to the
value given in (9).
In the even case, the values we obtain for and b are shown in Table 2 for four values of a .
Comparing with the values displayed in Table 1 one observes a better convergence of b to the
best value (9) whereas r remains unchanged compared to (10). As for the best estimate of , it
is:
best = 0.649561773880 1012 ,
(16)
that is to say:
1best = 1.539499459808 1012 .
(17)
We have proceeded similarly to determine the Ising-like subcritical exponent values displayed
in Table 3.
531
Table 3
Best estimates of the six first subcritical exponents for the Ising-like scalar model (i.e.,
even case), all digits are significant
1
2
3
0.655745939193
3.180006512059
5.912230612
4
5
6
8.796092825
11.79808766
14.896053176
Table 4
Best estimates of the odd-case subcritical exponents other than 1 for the scalar model
2
4.524390734
7.337650643
10.28390073
In the odd case, we obtain:

1 = 1.886703838091 1012 .
(18)
Table 4 displays the values of the other subcritical exponents of the same family as but with
a lower accuracy. Of course, the values presently obtained are in agreement with the previous
estimates [21,25].
2.2. Litims flow equation for the scalar field
Following a conjecture first stated in [18,19], the equivalence in the LPA between the Wilson
Polchinski flow (1) and the Litim optimized ERGE [17] for the running effective action [, t]
has been proven by Morris [20]. The Litim flow equation for the potential V (, t) reads in three
dimensions (compared to [20] an unimportant shift V V 1/3 is performed):
1
V + 3V .

1+V
2
It is related to (1) via the following Legendre transformation:

[ 12 2 U (, t)] + [ 12 2 + V (, t)] =
.
= U (, t)
V = 1
(19)
(20)
The general solution of the fixed point equation (V = 0) involves the following moving singularity (V is singular) at the arbitrary point 0 :
1
4
Vsing () = + |0 |3/2 .
3 3 0
(21)
2.2.1. Fixed point solution

The numerical study of the fixed point solution of (19) follows the lines described in the
preceding sections. This may be done independently, but due to (20), one may already deduce
from the previous study the expected results. Similarly to (5), the asymptotic behaviour of the
non-trivial fixed point potential is characterized by the integration constant bL in the following
expression [deduced from (19)]:
Vasy () = bL 6
12
1
1
1
.
+
O
+
2 8
3 150bL 4 6300bL
(22)
532
we are looking for is related to b as

It is easy to show from (5) and (20) that the value bL
follows:

1 5 5
bL = 6 ,
6 b
then, from the previous result (9) we get:
bL
0.00100025.
(23)
Similarly for the potential parameters

kL = V (0),
rL = V (0),
, they are related to the WilsonPolchinski counterparts k and r as
which correspond to bL
follows:
kL = k ,
r
rL =
.
1 r
(24)
(25)
This latter relation, using (10), gives:

rL 0.186064249470314 1015 .
(26)
As precedingly, those values do not provide the potential function V () the knowledge of
which requires an explicit numerical integration.
2.2.2. Eigenvalue equation
A linearization of the flow equation (19) near the fixed point solution V ():
V (, t) = V () + et h(),
provides the Litim eigenvalue equation:
(3 )h
h
h
= 0.
+
2
(1 + V )2
(27)
Taking into account (22), one can show that (27) admits a regular solution which, for ,
has the form:

1
2(3)
hasy () = S1
1 ( 3)(2 5)
2 10
2250bL

1
,
+ O 18
(28)
3
47 250bL 14
is given by (23). In the following we may set S = 1 since the normalization of the
in which bL
1
eigenfunction may be chosen at will.
As precedingly, we must distinguish between the odd and even eigenfunctions h(). The
shooting method gives the same values as in the WilsonPolchinski case (see [18,21,26,27]) and
we do not present them again.
533
3. Expansion in powers of the field

In advanced studies of the derivative expansion [28] or other efficient approximations of the
ERGE [29] and in the consideration of complex systems via the ERGEs [30], a supplementary
truncation in powers of the field is currently used (see also [8]). With a scalar field, this expansion
transforms the partial differential flow equations into ODEs whereas the fixed point or eigenvalue
ODEs are transformed into algebraic equations. Provided auxiliary conditions are chosen, the
latter equations are easy to solve analytically using a symbolic computation software. Actually
the auxiliary conditions currently chosen are extremely simple: they consist in setting equal to
zero the highest terms of the expansion so as to get a balanced system of equations.
A first kind of expansion, about the zero field, referred to as the expansion I in the following,
has been proposed by Margaritis et al. [11] and applied to the LPA of WegnerHoughtons ERGE
[6] (the hard cutoff version of the WilsonPolchinski equation). A second kind of expansion,
relative to the (running) minimum of the potential (expansion II), has been proposed by Tetradis
and Wetterich [12] and more particularly presented by Alford [13] using it, again, with the sharp
cutoff version of the ERGE.
It is known that, for the WegnerHoughton equation in the LPA, expansion I does not converge due to the presence of singularities in the complex plane of the expansion variable [31].
Expansions I and II have been more concretely studied and compared to each other by Aoki et
al. in [14] who also propose a variant to II (expansion III) by letting the expansion point adjustable. They showed, again on the LPA of the WegnerHoughton equation, that expansion II
is much more efficient than expansion I although it finally does not converge and expansion III
is the most efficient one. Expansions II and III work well also on the ERGE expressed on the
running effective action (effective average action, see the review by Berges et al. in [8]). The
convergence of those expansions have also been studied in [26] according to the regularization
scheme chosen and in particular for the Litim equation (19). In this latter study it is concluded
that both expansions I and II seem to converge although II converges faster than I.
A striking fact emerges from those studies, the WilsonPolchinski equation in the LPA, the
simplest equation, is never studied using the field expansion method. The reason is simple: none
of the expansions currently used works in that case.
Actually the strategy of these methods, which consists in arbitrarily setting equal to zero one
coefficient for the expansion I and two for the expansions II and III, is probably too simple. With
regards to this kind of auxiliary conditions, the failure observed with the WilsonPolchinski
equation is not surprising and, most certainly, there should be many other circumstances where
such simple auxiliary conditions would not solve correctly the derivative expansion of an ERGE.
In the following sections we examine two alternative methods with more sophisticated auxiliary conditions. We show that they yield the correct solution for the WilsonPolchinski and its
Legendre transformed (Litim) equations. Both methods are associated to expansion I (about the
zero-field). The first one has recently been proposed in [15] as a method to treat the two point
boundary value problem of ODEs. It relies upon an efficient account for the large field behaviour
of the solution looked for. An attempt of accounting for this kind of behaviour within the field
expansion had already been done by Tetradis and Wetterich via their Eq. (7.11) of [12]. In the
present work, a much more sophisticated procedure is used. It relies upon the construction of an
added auxiliary differential equation (ADE). We refer to it in the following as the ADE method.
The second method is new. It relies upon the approximation of the solution looked for by a generalized hypergeometric function. We refer to it in the following as the hypergeometric function
approximation (HFA) method.
534
4. Auxiliary differential equation method

Let us first illustrate the auxiliary differential equation (ADE) method on the search for the
non-trivial fixed point in the LPA for both the WilsonPolchinski equation (2) and the Litim
optimized equation (19). Since there are two boundaries (the origin and the point at infinity),
we distinguish between two strategies.
An expansion about the origin in the equations (small field expansion) and the account for the
leading high field behaviour of the regular solution which we are looking for. This determines
the value of r or rL .
A change of variable 1/ or 1/ which reverses the problem: an expansion about
infinity (new origin) in the equations (high field expansion) and the account for the leading
small field behaviour of the regular solution which we are looking for. This determines the
.
value of b or bL
4.1. WilsonPolchinskis fixed point
4.1.1. Small field expansion and leading high field behaviour
For practical and custom reasons,2 instead of (2) we consider the equation satisfied by the
function w(x) related to the derivative of the potential U () as follows:

U () = w 2 ,
(29)
so that, with x = 2 , the fixed point equation (2) reads:
4xw 2w 2 4xww + (6 x)w + 2w = 0,
(30)
in which a prime indicates a derivative with respect to x.

This second order ODE has a singular point at the origin and, by analyticity requirement, the
solution we are looking for depends on a single unknown integration-constant (noted r below).
Let us first introduce the expansion I of Margaritis et al. [11]. The function w(x) is expanded
up to order M in powers of x:
wM (x) = r +
an x n ,
(31)
n=1
and inserted into the fixed point equation (30).

Requiring that (30) be satisfied order by order in powers of x provides an unbalanced system
of M algebraic equations with M + 1 unknown quantities {r, a1 , . . . , aM } [Eq. (30) is then satisfied up to order M 1 in powers of x]. With a view to balancing the system, aM is simply set
as M grows, then it constitutes the
equal to zero and if the solution involves a stable value rM
estimate at order M of the fixed point location corresponding to expansion I. As already mentioned, in the case of the WilsonPolchinski equation (30) under study, the method fails: all the
are positive whatever the value of M whereas the correct value should be
values obtained for rM
negative as shown in Section 2.1.1.
2 The change x = 2 is useful in practice to avoid some degeneracies observed in [15] when forming the auxiliary
differential equation. Taking the derivative f = U is only a question of habit.
535
In the ADE method, the condition aM = 0 is not imposed. The previous algebraic system
is first solved in terms of the unknown parameter r so as to get the generic solution of (30) at
order M in powers of x:
wM (r; x) = r +
an (r)x n .
(32)
n=1
In order to get a definite value for r, instead of arbitrarily imposing aM (r) = 0, an auxiliary
condition is formed which explicitly accounts for the behaviour at large given by (5). With
w(x), this behaviour corresponds to:
wasy (x) = 1,
(33)

wasy
(x) = 0.
(34)
x
x
The auxiliary condition is obtained via the introduction of an auxiliary differential equation:
Consider a first order differential equation for w(x) constructed as a polynomial of degree s
(eventually incomplete) in powers of the pair (w, w ):
G1 + G2 w + G3 w + G4 w 2 + G5 ww + G6 w 2 + + Gn w sq w q = 0,
(35)
in which, when the degree s of the polynomial is saturated then q = s and the number n of
coefficients Gi is equal to (s + 1)(s + 2)/2, conversely when it is not then 0 q < s and
n = s(s + 1)/2 + q + 1.
The constant coefficients Gi are then determined as functions of r by imposing that the
solution wM (r; x) of (30) previously determined for arbitrary r at order M in powers of x
be also solution of (35) (at the same order M). Due to an arbitrary normalization which
allows to fix, for example G1 = 1, a simple counting shows that the identification implies
M = n 1. The resulting set {Gi (r); i = 2, . . . , n} is formed of rational functions of the
unknown parameter r. Hence, a new differential equation for w(x) is obtained:
1 + G2 (r)w + G3 (r)w + G4 (r)w 2 + G5 (r)ww
+ G6 (r)w 2 + + Gn (r)w sq w q = 0,
(36)
which is satisfied by construction at order M in powers of x by (32) which is already solution

at the same order of (30).
The last step is then to impose that the new equation (36) be also satisfied when x .
Taking into account (33), (34) it comes the final auxiliary condition:
1 + G2 (r) + G4 (r) + + Gs(s+1)/2+1 (r) = 0.
(37)
Solving this auxiliary condition for r amounts to determining the roots of a polynomial in r.
As the order M grows some root values appear to be stable. Those stable values are candidates
for the fixed point solutions we are looking for. In a way similar to [16], the obtention of the
auxiliary condition may be obtained without determinating explicitly the coefficient functions
Gi (r). For this, it is sufficient to consider the matrix F of the homogeneous system of linear
equations for all the Gi s formed with Eq. (35) to which is added its expression when x .
When the function w(x) is replaced by the expansion (32) at the required order the matrix F
depends only on the coefficients an (r) of the Taylor expansion (32) and the auxiliary condition
536
Fig. 1. Distribution of the real roots rM (open circles) of (37) as function of the order M of the Taylor series about
the origin (31) [with the ADE pair (w, xw )]. A threefold accumulation occurs about the expected fixed points: trivial high temperature (r = 1) and Gaussian (r = 0) fixed points and about the non-trivial WilsonFisher fixed point
(r 0.2286) [LPA, d = 3].
then finally reduces to:

det F = 0.
(38)
Before going further, it is worthwhile indicating that a variant of the method which consists in
replacing w by xw in the auxiliary differential equation (35) has appeared more efficient [e.g.,
see Fig. 2].
Fig. 1 shows the distribution of all the real roots rM of (37) for the variant as the order M varies
up to 28. The three expected fixed points encountered in Section 2.1.1 are clearly evidenced
by a threefold accumulation about the respective values 1 (HT), 0 (Gaussian) and r (Wilson
Fisher). Although a huge accumulation of roots around the right value occurs, the approach to
r , which we are interested in, may be followed step by step as the order M grows.
4.1.1.1. Selection of the root To select the right value of the root corresponding to the nontrivial WilsonFisher fixed point, the following procedure has been applied. We know that the
root of interest is negative and real, then we select the first negative real root that appears at
the smallest possible order. At the next order we choose the real root the closest to the previous
choice and so on. We obtain this way with M = 28 the following excellent estimate:
r = 0.22859820243702,
(39)
which coincides, up to the 14th digit, with the estimate (10) obtained by the shooting method.
Fig. 2 shows the accuracy obtained on r by selecting the roots this way as M varies.
4.1.2. Subleading high field behaviour
Eqs. (33), (34) used in the preceding calculations express exclusively the limit of the solution
w (x) when x , and we get the unique condition (37) to estimate r . In fact there are
537
Fig. 2. Approximate number of accurate digits Nd = log |1 rM /r | obtained on the selected roots rM as a function
of M and for two ADE pairs; the original (w, w ): crosses, and the variant (w, xw ): open circles [r is given in Eq. (10)].
The highest values of M in each case is limited by time computing. A better efficiency is obtained with the variant.
higher correction terms to (33), (34) which vanish as x [the first of which correspond
to those written in (5)]. Such subleading contributions may as well be imposed in (36). In so
doing, we require the auxiliary differential equation to be satisfied not only at infinity but also
in approaching this point. Consequently we obtain several auxiliary conditions similar to (37),
each of them corresponding to the cancellation of the coefficient of a given power of x. We have
used them to determine r again (the asymptotic constant b factorizes in the first subleading
conditions so obtained). The results are similar to those obtained precedently with the leading
conditions (33), (34) alone. We have observed only a slight decrease in the accuracy: the higher
the subleading term considered the weaker the convergence to r . This shows the coherence of
the ADE method: the auxiliary condition is not an isolated point condition, it emanates from a
differential equation constructed to be satisfied by the function looked for.
When the order of the subleading contribution is high enough, the constant b no longer factorizes and the subleading auxiliary condition depends non-trivially on the (non-independent)
integration constants (r and b) characterizing the fixed point solution. We have tried to determine
the value b by imposing the individual vanishing of such contributions for r = r . Unfortunately,
at the orders considered, the only knowledge of r suffices to satisfy the condition (whatever the
value of b). It is possible that considering much higher orders would allow us to get an estimate
of b this way.
4.1.3. High field expansion and leading small field behaviour
With the determination of b by the ADE method in view, let us perform the change of variable
x y 5 and the following change of function:

u(y) = y 2 w y 5 1 ,
(40)
so that, from (5) and (29), u (y) has the following form for small y:

1
1
u (y) = A + A 2 y 2 A 3 y 4 + O y 5 ,
5
25
(41)
538
Fig. 3. In the WilsonPolchinski case, A does not converge (dots) to the right value 2.78195 (horizontal line) but to
2.73532.
with
A = 6b /5.
The fixed point differential equation (2) is then transformed into:

10yu2 + 5 5 + 2y 5 u 4 y 4 5y 2 u u + 4y 6 u = 0,
(42)
(43)
the solution of which must satisfy the following condition, see (41):
u (0) = A ,
u (0) = 0,
with A to be determined so as, using (31), (40), to get at infinity:
uasy (y) = 0.
y
The ADE method described in the preceding sections is used to determine the value of A .
Since there are some holes in the first terms of the series (41), the first significant estimates are
obtained for values of M higher than in Section 4.1.1. Fig. 3 shows that the selected sequence
of roots corresponding to A converges to 2.73532 whereas, according to (9), (42), the right
value expected from the shooting method is 2.78195.
This failure of the ADE method in determining correctly A is presumably due to the zero
radius of convergence of the Taylor series of u (y) about y = 0. Actually, we have estimated
this radius as the limit of the ratio of two consecutive terms and observed that it goes slowly
but continuously to zero as the order M increases. This contrasts with the case of w(x) for
which the same procedure quickly tends to the following finite limit for the fixed point solution
corresponding to (10), (39):
RWP = 5.72167.
(44)
539
Notice that, although the ADE method does not provide the right estimate of A (or b ), it
gives a value close enough to it to be used as a guess in the shooting method.
4.2. Litims fixed point
4.2.1. Small field expansion and leading high field behaviour
For convenience we perform the following change, compared to Section 2.2:
1
V () = w 2 ,
3
(45)
so that the fixed point equation corresponding to (19) reads (with x = 2 ):

1
(46)
= 0.
+ 4x w
The singularity at x = 0 of this second order ODE allows us to look for an analytic solution
the following conditions at the origin:
which satisfies, in terms of a single unknown parameter k,
3w x w
1 + 2w
w(0)
= k,
1
1
w (0) =
,
6k 2
with k adjusted to k so as to reach at infinity [from (22)]:
3
w asy
(x)
= bL
x +

1
1
+ O x 6 .
2
4
150bL x
6300bL x
(47)
(48)
(49)
The expected value of k is related to rL given in (26) as:

k =
1
.
3(1 + rL )
It is also related to k given in (11) via (24), (45) as k = k + 1/3. Consequently the estimation by the shooting method is:
k = 0.409532734145674 1015 .
(50)
The object of this section is thus to test whether the ADE method yields that value of k [and
given in (23)].
also that of bL
Contrary to the WilsonPolchinski case, the asymptotic behaviour (49) does not reach a finite
is still supposed unvalue when x . But the third derivative of w does. Hence, since bL
known, the auxiliary first order differential equation (35) may be used with w and w replaced
x n ). Actually, both of these two
respectively by w (4) and w (5) (where w (n) stands for dn w/d
derivatives go to zero as x so that finally the auxiliary condition similar to (37), but with
another normalization of the Gi s (e.g., G2 = 1), reduces to:
= 0,
G1 (k)
(51)
whereas the function w(

x)
is expanded up to order M in powers of x and inserted into (46) to
get the solution at this order as function of k:

x)
= k +
w M (k;
n=1
x n .
an (k)
(52)
540
Similarly to the WilsonPolchinski case, the complete set of real roots of (51) shows accumulations about the expected fixed point values. However the selection process described previously
fails in picking the right value k (of the non-trivial fixed point) although it is present among
the roots. Actually, for M = 14 the selection gives 0.40962781972971 whereas a better value
(0.40953273321235) exists at the same order [compare with (50)]. The variant utilized in the
preceding case which consists in replacing w (5) by x w (5) does not circumvents this difficulty.
If instead of (w (4) , w (5) ) as ADE pair, we consider the combination h = 3w x w and its
to save some time computing), then the new
derivative h with respect to x (or the variant xh
pair, according to (49), vanishes also as x , and we observe, this time, that the selection
process works again. This way, at order M = 19 the selection gives:
k = 0.40953273416,
a value which coincides with (50) up to the 10th digit. No doubt that considering higher values
of M would have improved the accuracy. We note that, as with WilsonPolchinskis function,
the radius of convergence of the Taylor series of w(
x)
about the origin is finite, and is about:
RL 11.5.
(53)
Let us specify however that, contrary to the WilsonPolchinski case, the test of the ratio
ai /ai+1 of two consecutive terms of the Taylor series about the origin does not converge. We
have obtained (53) by explicitly performing a partial summation of the series and studying it as
a function of x.
Nevertheless, we have also observed that the ratio |ai /ai+3 | raised to the power
1/3, roughly converges to (53). This remark will have some importance in Section 5.4.
Since expansions I and II work in the Litim case (see [26]), we can compare the ADE method
with those two methods. Fig. 4 shows the respective accuracies obtained on k with the three
methods as functions of the order M of the field expansion. One sees that expansion II and the
Fig. 4. Approximate number of accurate digits Nd = log|1 kM /k | [with k given by (50)] as functions of M for the
estimations of the Litim fixed point value of k using three methods: expansion I (black dots), expansion II (crosses), and
ADE (open circles). A better efficiency is obtained with the ADE method
541
ADE method provide better results than expansion I (which likely does not converge) and that
the ADE method is most efficient than expansion II (we have not studied expansion III).
4.2.2. Subleading high field behaviour
As in the case of WilsonPolchinskis equation, the subleading terms in (49) may be used to
impose the auxiliary condition not only at infinity but also in approaching this point whatever the
value of x.
We observe the same phenomenon as in Section 4.1.2: the higher the subleading term
cannot be determined by imposing the
considered the weaker the convergence to k whereas bL
individual vanishing of the subleading contributions for k = k .
However, the fact that the asymptotic behaviour (49) is an integer power of x provides us
from the knowledge of k as a boundary limit (a point
with the opportunity of determining bL
(3)
condition). Actually, since w 6bL when x , we may choose (w (3) , w (4) ) as ADE pair
the resulting auxiliary condition
[or the variant (w (3) , x w (4) )], and for k fixed to k solve for bL
at infinity. The accuracy on bL obtained this way is not as large as in the case of k , nevertheless,
for M = 31 we obtain the following estimation:
0.001007,
bL
(54)
which is rather close to the shooting value (23). We indicate also that rough estimates of bL
already sufficiently accurate to be used as guesses in the shooting method are obtained for small
values of M, e.g., 0.000989 for M = 11 or even 0.0012 for M = 5.
4.2.3. High field expansion and leading small field behaviour

directly by the ADE method, we invert the boundaries by
With a view to determining bL
1
changing the variable x y and by performing the following change of function:

1
u(
y)
= y 3 w
(55)
,
y
so that, from (49), we deduce that u (y)
has the following form for small y:
= bL
+
u (y)

y 5
y 7
+ O y 9 .
2
150bL
6300bL
(56)
The differential equation for u(

y)
is:

y 4 + 18y(
u )2 u 30u + y 2 (1 + 4u ) = 0.
The solution must satisfy the following condition at the origin y = 0 [see (56)]:
,
u (0) = bL
u (0) = 0,
to be determined so as, using (47), (48), (55), to get at infinity:
with bL

1
1 2
= k y +
y + O(y).
u asy (y)
6k 2
. For this we
As previously, we use the ADE method with a view to determining the value of bL
(4)
(5)
consider, the pair (u , u ) which vanishes at infinity (y ). Since there are some holes in
the first terms of the series about the origin, see (56), the first significant estimates are obtained for
542
values of M higher than with the original function w(

x).
Although the positive roots obtained for
is positive) have the right order of magnitude compared to (23) the apparent
bL (we know that bL
convergent sequences do not provide the right value. Again, as in the WilsonPolchinski case,
we think that the failure of the ADE method is due to the (observed) zero radius of convergence
of the Taylor series for u(
y)
about the origin.
4.3. Eigenvalue estimates
Let us consider the eigenvalue problem with the ADE method. This time two coupled nonlinear ODEs have to be solved together (the fixed point equation and the linearization of the flow
in the vicinity of the fixed point). We can solve these two equations together as the order of the
field expansion M grows or consider separately the eigenvalue equation after having solved the
fixed point equation with some accuracy. With the aim to be short, we present only the latter
possibility which illustrates well the property of convergence of the method.
4.3.1. WilsonPolchinskis eigenvalues
4.3.1.1. Small field expansion and leading high field behaviour Using a change of eigenfunction, g v, similar to (29) for the fixed point function, it comes:
in the even case:

g () = v 2 ,
and Eq. (13) yields the following behaviour at large x = 2 :

vasy (x) =

2(3 )
S0 x (2+)/5 1 + O x 2/5 ;
5
in the odd case:

g () = v 2 ,
and Eq. (13) gives:

vasy (x) =

2(3 )
S0 x (12)/10 1 + O x 2/5 .
5
The arbitrariness of the global normalization of the eigenfunctions allows to choose v(0) = 1
(even) and v (0) = 1 (odd) corresponding respectively to some definite values of S0 . So defined,
the functions v(x) and v (x) vanish at infinity provided that > 2 in the even case and >
1/2 in the odd case. Hence one could expect that, with the simple condition at infinity: v =
v = 0 imposed in the auxiliary differential equation, the ADE procedure will, at best, allow the
determination of exclusively the leading (1 = 1/) and first subleading (2 = 1 ) eigenvalues
in the even case and of only the trivial eigenvalue 1 = 1 in the odd case [see the values
of these quantities in Eqs. (17), (18) and Tables 3, 4]. Actually it is better than that since, as
M grows, we observe among the real roots of the auxiliary condition for that a hierarchy
of successive accumulations takes place about the right values of the leading and subsequent
eigenvalues [see Fig. 5].
Within each of these accumulations of real roots, we have been able to follow without ambiguity a convergent sequence to the right estimate. At order M = 20 with the ADE pair (v, v )
543
Fig. 5. Accumulations of real roots (open circles) of the auxiliary condition (37) about eigenvalues as the order M of
the series varies in the WilsonPolchinski even case. From top to bottom: 1 = 1/ (second horizontal line), 2 = 1
(third horizontal line) and 3 = 2 (fourth horizontal line). A simple criterion of choice allows to determine their
estimates at M = 20, see the values in Eqs. (57)(59). An accumulation also occurs about the spurious value 5.8 (first
horizontal line).
supposed to vanish at infinity, and r fixed to the value given in (39), we have obtained the
following estimates in the even case
= 0.64956177386,
(57)
1 = 0.65574592,
(58)
2 = 3.178,
(59)
where the number of digits has been truncated with regard to the accuracy of the estimates obtained [by comparison with (16) and Table 3]. We see that the accuracy decreases as the order of
the eigenvalue grows but also that we obtain an estimate of 2 whereas for that value v does not
vanish at infinity.
The same kind of observations stands in the odd case. We take the opportunity to indicate

that choosing the ADE pair (f, f ) with f = (12)
10 v xv makes f vanish for > 3/2 and
the procedure gives a better accuracy on 1 than with the pair (v, v ). This way we obtain the
following estimate [at order M = 20, compare with (18)]
1 = 1.886718.
We have also noted the presence of accumulations of real roots about spurious positive values
of order 5.8 in the even case and 3.77 in the odd case.
4.3.2. Litims eigenvalues
The determination using the ADE method of the eigenvalues from the Litim flow equation
follows the same lines as previously for the WilsonPolchinski flow equation. We limit ourselves
in this section to a brief presentation of the main differences encountered.
544
4.3.2.1. Small field expansion and leading high field behaviour Compared to (27), we perform
a change of eigenfunction, h vL , according to the symmetry considered:
in the even case:

h() = vL 2 ,
then Eq. (28) yields the following behaviour at large x = 2 :

= S1 x (3) 1 + O x 5 ;
vL asy (x)
in the odd case:

h() = vL 2 ,
and Eq. (28) gives:

vL asy (x)
= S1 x (5/2) 1 + O x 5 .
and vL (x)

vanish at infinity provided that > 3 in the
So defined, the two functions vL (x)
even case and > 5/2 in the odd case (whereas the arbitrary global normalization of the eigenfunctions allows to choose vL (0) = 1 (even) and vL (0) = 1 (odd) corresponding respectively to
specific values of S1 ).
Although it works, the original ADE pair (vL , vL ) is not the most efficient choice to obtain
estimates of the first non-trivial eigenvalues. A better choice appears to be the pairs (f (x),
f (x))

with f (x)
= (3 )vL (x)
xv
L (x)
in the even case and f (x)
= (5/2 )vL (x)
xv
L (x)
in
the odd case (they correspond to eigenfunctions which vanish as x for more negative
values of ). With these choices and k = 0.4095327341457 we identify immediately the trivial
eigenvalues 0 = 3 in the even case and 1 = 2.5, 2 = 0.5 in the odd case but also, for M = 20,
we obtain good estimates of the non-trivial leading and first subleading eigenvalues:
= 0.649561774,
1 = 1.8867037,
1 = 0.6557455,
2 = 3.180008,
3 = 5.896,
2 = 4.5241,
where the numbers of digits have been limited with respect to the estimated accuracy [compare
with (16), Table 3 (even) and (18), Table 4 (odd)]. For each eigenvalue, the successive estimates
may be followed unambiguously step by step when M grows so that the right values may be
easily selected following the rules defined precedently.
We notice also the presence of spurious convergences and especially in the even case to the
value about 5.8 already encountered with the WilsonPolchinski case.
5. Approximating by hypergeometric functions (HFA)
The ADE method is most certainly efficient in many cases but it is relatively heavy regarding
the computing time whereas the current methods, when they work, are lighter. In addition, none
of these methods provides a global solution to the ODE studied: they yield an approximate value
of the integration constant but not a function as global approximation of the solution looked for.
We propose in this section an alternative method which is lighter than the ADE method and
which provides a global approximation of the solution of interest. This new method is based
on the definition property of the generalized hypergeometric functions. Let us first review the
definition and main properties of these functions.
545
5.1. Generalized hypergeometric functions

n
For x C, a series S =
n=0 an x is hypergeometric (see for example [32]) if the ratio
an+1 /an is a rational function of n, i.e.,
P (n)
an+1
=
,
an
Q(n)
for some polynomials P (n) and Q(n).
If we factorize the polynomials, we can write:
(n + 1 )(n + 2 ) (n + p )
an+1
= 0
.
an
(n + 1 )(n + 2 ) (n + q )(n + 1)
(60)
The factor (n + 1) in the denominator may or may not result from the factorization. If not, we
add it along with the compensating factor in the numerator. Usually, the global factor 0 is set
equal to 1.
If the set {i } includes negative integers, then S degenerates into a polynomial in x.
When it is not a polynomial, the series S converges absolutely for all x if p q and for
|x| < 1/|0 | if p = q + 1. It diverges for all x
= 0 if p > q + 1.
The analytic continuation of the hypergeometric series S with a non-zero radius of convergence is called a generalized hypergeometric function and is noted:
1
S.
a0
p Fq (x) is a solution of the following differential equation (for 0 = 1):

( + 1 1) ( + q 1) x( + 1 ) ( + p ) p Fq (x) = 0,
p Fq (1 , . . . , p ; 1 , . . . , q ; 0 x) =
(61)
where
d
.
dx
When p > 2 or q > 1, the differential equation (61) is of order max(p, q + 1) > 2. It is of
second order when q = 1 and p = 0, 1 or 2. It is of first order when q = 0 and p = 1.
2 F1 is currently named the hypergeometric function. A number of generalized hypergeometric
functions have also special names: 0 F1 is called confluent hypergeometric limit function and 1 F1
confluent hypergeometric function.
In the cases p q for fixed {i } and {i }, p Fq (x) is an entire function of x and has only one
(essential) singular point at x = .
For p = q + 1 and fixed {i } and {i } in non-polynomial cases, p Fq (x) does not have pole nor
essential singularity. It is a single-valued function on the x-plane cut along the interval [1, ],
i.e., it has two branch points at x = 1 and at x = .
Considered as a function of {i ; i = 1, . . . , q}, p Fq (x) has an infinite set of singular points:
=x
(1) i = m, m N which are simple poles;

(2) i = which is an essential singular point (the point of accumulation of the poles).
As a function of {i ; i = 1, . . . , p}, p Fq (x) has one essential singularity at each i = .
The elementary functions and several other important functions in mathematics and physics
are expressible in terms of hypergeometric functions (for more detail see [32]).
546
The wide spread of this family of functions suggests trying to represent the solution of the
ODEs presently of interest in this article, under the form of a generalized hypergeometric function.
5.2. The HFA method
For the sake of the introduction of the new method, let us first consider the WilsonPolchinski
fixed point equation (30) and the truncated expansion (32) in which the coefficients an (r)
(n = 1, . . . , M) are already determined as function of r via a generic solution of (30) truncated
at order M (in powers of x). The question is again to construct an auxiliary condition to be
imposed with a view to determining the fixed point value r . To this end, by analogy with the
generalized hypergeometric property definition recalled in Section 5.1, we construct the ratio of
two polynomials in n:
m1
i1
Pm1 (n)
i=1 ci n
(62)
,
= m
2
i1
Qm2 (n)
i=1 di n
so that Pm1 (n)/Qm2 (n) match the M 2 ratios an+1 (r)/an (r) for n = 1, . . . , M 2. Hence,
accounting for the arbitrariness of the global normalization of (62), the complete determination of
the two sets of coefficients {ci ; i = 1, . . . , m1 } and {di ; i = 1, . . . , m2 } as functions of r implies
m1 + m2 = M 1. Finally, the auxiliary condition on r is obtained by requiring that the last
(still unused) ratio aM (r)/aM1 (r) satisfies again the n-dependency satisfied by its predecessors,
namely that:
m1
i1
aM (r)
i=1 ci (r)(M 1)
(63)
=
.
m
2
i1
aM1 (r)
i=1 di (r)(M 1)
The auxiliary condition so obtained is a polynomial in r, the roots of which are candidates to
). Notice that, to obtain faster this auxiliary
give an estimate at order M of r (noted below rM
condition, one may avoid the calculation of the coefficients ci (r) and di (r) by following the same
considerations as those leading to (38) with the ADE method.
At this point, the method potentially reaches the same goal as the ADE and other preceding
methods. However, according to Section 5.1, in determining the ratio of polynomials (62) we
have also explicitly constructed the function
FM (x) = rM
m1 +1 Fm2 (1 , . . . , m1 , 1; 1 , . . . , m2 ; 0 x),
(64)
is the selected estimate of r , the sets { } and { } are the roots of the two
in which rM
i
i
whereas:
polynomials Pm1 (n) and Qm2 (n) when r = rM
0 =
)
cm1 (rM
).
dm2 (rM
(65)
Now, by construction, FM (x), has the same truncated series in x as the solution of (30) we are
looking for. This function is thus a candidate for an approximate representation of this solution.
It is worth noticing that, contrary to the ADE method, the HFA method does not make an
explicit use of the conditions at infinity (large x) to determine r . Only a local information, in
the neighbourhood of the origin x = 0, is explicitly employed.
Let us apply the method to the two equations of interest in this paper.
547
with the HFA method

Fig. 6. Respective approximate number of digits (defined in the caption of Fig. 2) obtained for rM
(crosses) and the ADE method (open circles) for the WilsonPolchinski fixed point equation. Whereas at a given order M
the accuracy is similar, a smaller time computing is necessary with the HFA method.
5.3. WilsonPolchinskis equation

5.3.1. Fixed point
We know that the absolute value of the ratio an (r )/an+1 (r ) has a definite value RWP [given
by Eq. (44)] as n . Consequently, we must consider the ratio (62) with m1 = m2 (this
implies also that M be odd). In this circumstance, according to Section 5.1, the relevant hypergeometric functions have a branch cut on the positive real axis (as functions of 0 x). Consequently
the analytic continuation to large positive values of x is only possible if 0 < 0. We note also
that, according to (44), |0 | should converge to 1/RWP = 0.174774. Finally by considering the
large x behaviour directly on (61), it is easy to convince oneself that the leading power is given
by one of the parameters {i }, consequently we expect to observe a stable convergent value
among the i s toward the opposite of the leading power at large x of the solution looked for.
For this reason, instead of the function w(x) of Section 4.1.1 the limit of which is 1 as x
[see (33)], we have considered the translated function wt (x) = w(x) 1 which, according to
Eqs. (5) and (29), (42), tends to A x 2/5 . In this case we thus expect to observe a stable value
among the i s about 0.4 with the eventual possibility of estimating A .
When looking at the roots of the auxiliary condition (63) as M varies, we obtain the same
kind of accumulations about the expected fixed point value r as shown in Fig. 1 (with much
less points however). We can also easily select the right non-trivial solution using the procedure described just above (39). We get precisely this excellent estimate with M = 25 and a
reduced computing time compare to the ADE method. Fig. 6 shows the accuracies obtained
on r (crosses) compared to the ADE method (open circles).
Furthermore, the sets of parameters of the successive hypergeometric functions involve two
stable quantities the values of which at M = 25 are:
0 = 0.174775,
(66)
548
1 = 0.3962.
(67)
Those two results are quantitatively and qualitatively very close to the expected values (respectively 0.174774 and 0.4 as given just above).
This clearly shows that the hypergeometric function determined this way provides us with a
really correct (but approximate) global representation of the fixed point function. This contrasts
strongly with the numerical integration of the ODE which, due to the presence of the moving
singularity, never provides us with such an approximate global representation of the solution
looked for.
From (67) we have obtained a rough estimate of A (= 6b /5) by a direct consideration of
the value of the corresponding function FM (x) defined in (64) for some relatively large value of
x and we obtain A 2.6 what is a sufficiently accurate estimate to serve as a guess in the
shooting method.
We have also tried to determine, using the HFA method, the value A directly from the reverse side corresponding to (43). We have not improved the previous biased estimate obtained
by ADE (about A = 2.735). We do not understand the significance of this coincidence. We
recall, however, that the radius of convergence of the Taylor series of u (y) about y = 0 probably
vanishes. This biased result shows again that the property of convergence of the Taylor series is
crucial for the accuracy of the two methods.
5.3.2. Eigenvalues
We have also applied the HFA method to the determination of the eigenvalues. With M = 17,
we have easily and without ambiguity obtained the following excellent estimates [compare with
(16), (18) and Tables 3 and 4]:
= 0.649561774,
3 = 5.9122294,
1 = 1.886703839,
1 = 0.6557459393,
4 = 8.796045,
2 = 4.5243903,
2 = 3.18000653,
5 = 11.8004,
3 = 7.337635.
These results show a greater efficiency than with the ADE method especially in the determination of the subleading eigenvalues.
It is worth indicating also that, surprisingly enough, we observe again (i.e., as with the ADE
method) the presence of convergences to the same spurious eigenvalues: 5.8 and 3.8 in the even
and odd cases respectively.
5.4. Litims equation
5.4.1. Fixed point
provides again
Applying the HFA method with the ratio of two successive coefficients an (k)
an accumulation of roots about the right value of k given in (50). However, this time, we have
encountered some difficulties in defining a process of selection of the right root. We obtain the
following estimate for M = 21:
k 0.409531,
which is not bad [compare with (50)] but not as satisfactory as in the preceding Wilson
Polchinskis case.
With regard to the transformation (20) and the preceding success of the HFA method, it is not
amazing that the representation of the solution in the Litim case be more complicated than in the
WilsonPolchinski case.
549
We have already mentioned that, instead of the ratio of two successive terms of the series
it is a shifted ratio that roughly converges to the finite radius of convergence (53). As a
an (k),
matter of fact, if we use the ratios
an+3 (k)
,
an (k)
instead of the ratio an+1 /an without changing the procedure 3 described in Section 5.2, then we
get a better estimate for M = 21 [compare with (50)]:
k 0.409532737,
although the convergence properties are not substantially modified.
Because the case is apparently more complicated than precedently, we do not pursued further
the discussion of the global representation of the fixed point solution by generalized hypergeometric functions.
5.4.2. Eigenvalues
For the eigenvalue problem, a similar difficulty occurs where the right values do not appear as
clear convergent series of roots. At order M = 17, we get the following estimates:
1 = 0.6576,
= 0.64955,
1 = 1.89,
2 = 3.20,
3 = 5.8,
2 = 4.5.
As in the case of the fixed point determination, if instead of applying the method with the ratio
we consider the ratios
of two successive terms of the series an (k)
an+3 (k)
,
an (k)
then we get better estimates for M = 19:
= 0.649561774,
1 = 1.88671,
1 = 0.65575,
2 = 3.1807,
3 = 5.905,
2 = 4.524,
where the numbers of digits have been limited having regard to the estimated accuracies [compare with (16), Table 3 (even) and (18), Table 4 (odd)].
6. Summary and conclusions
We have presented the details of a highly accurate determination of the fixed point and the
eigenvalues for two equivalent ERGEs in the local potential approximation. First, we have made
use of a standard numerical (shooting) method to integrate the ODEs concerned. Beyond the test
of the equivalence between the two equations, already published in [21], the resulting numerics
have been used to concretely test the efficiency of two new approximate analytic methods for
solving two point boundary value problems of ODEs based on the expansion about the origin of
the solution looked for (field expansion).
3 Notice that the procedure does not define some generalized hypergeometric function of x 3 . This would have been
obtained by considering separately three series in the original series. Then a combination of three generalized hypergeometric functions would have represented the solution looked for.
550
We have considered explicitly those two methods applied to the study of the two equivalent
ODEs. We have shown that they yield estimates as accurate as those obtained with the shooting
method provided that the Taylor series about the origin of the function looked for has a non-zero
radius of convergence.
This is an important new result since, up to now, no such approximate analytical method was
known to work in the simplest case of the WilsonPolchinski equation. In the case of the Litim
equation the two methods converge better than the currently used expansions (usually referred to
as I and II in the literature, see, e.g., [14]). Our results support concretely the conclusions of [19]
which indicated that the high field contributions were important in the WilsonPolchinski case
whereas they were less important in the Litim case.
The first of the two methods relies upon the construction of an auxiliary differential equation
(ADE) satisfied by the Taylor series at the origin and to which is imposed the condition of the
second boundary (at infinity) [15].
The second method (HFA) is new. It consists in defining a global representation of the solution
of the ODE via a generalized hypergeometric function. The HFA method provides the advantage
of yielding a global (approximate) representation of the solution via an explicit hypergeometric
function.
In both cases it is possible to obtain easily (with few terms in the field expansion) rough
estimates of the solution which may be used as guesses in a subsequent shooting method.
The procedures may be applied to several coupled ODEs as shown in [15] for the ADE
method. Hence, we hope that the present work will make easier and more efficient future explicit (and ambitious) considerations of the derivative expansion of exact renormalization group
equations.
Acknowledgement
We thank D. Litim for comments on an earlier version of this article.
References
[1] M. Gell-Mann, F.E. Low, Phys. Rev. 95 (1954) 1300;
K.G. Wilson, Phys. Rev. 140 (1965) B445;
K.G. Wilson, Phys. Rev. D 2 (1970) 1438.
[2] K.G. Wilson, Rev. Mod. Phys. 47 (1975) 773.
[3] J. Zinn-Justin, Euclidean Field Theory and Critical Phenomena, fourth ed., Clarendon Press, Oxford, 2002;
A. Pelissetto, E. Vicari, Phys. Rep. 368 (2002) 549.
[4] C. Wetterich, Phys. Lett. B 301 (1993) 90;
M. Bonini, M. DAttanasio, G. Marchesini, Nucl. Phys. B 409 (1993) 441;
T.R. Morris, Int. J. Mod. Phys. A 9 (1994) 2411;
U. Ellwanger, Z. Phys. C 62 (1994) 503.
[5] K.G. Wilson, Irvine Conference, 1970, unpublished;
See the equation in K.G. Wilson, J.B. Kogut, Phys. Rep. 12 (1974) 75.
[6] F.J. Wegner, A. Houghton, Phys. Rev. A 8 (1973) 401.
[7] J. Polchinski, Nucl. Phys. B 231 (1984) 269.
[8] K.I. Aoki, Int. J. Mod. Phys. B 14 (2000) 1249;
C. Bagnuls, C. Bervillier, Phys. Rep. 348 (2001) 91;
J. Berges, N. Tetradis, C. Wetterich, Phys. Rep. 363 (2002) 223;
J. Polonyi, Cent. Eur. J. Phys. 1 (2003) 1;
For a recent overview of advanced functional RG methods, see J.M. Pawlowski, hep-th/0512261;
For a pedagogical introduction see B. Delamotte, cond-mat/0702365.
551
[9] J.F. Nicoll, T.S. Chang, Phys. Lett. A 62 (1977) 287.

[10] G.R. Golner, Phys. Rev. B 33 (1986) 7863.
[11] A. Margaritis, G. dor, A. Patks, Z. Phys. C 39 (1988) 109;
Actually the procedure followed there had already been implemented in another context by F.M. Fernandez, E.A.
Castro, J. Phys. A 14 (1981) L485;
J.R. Silva, S. Canuto, Phys. Lett. A 88 (1982) 282;
J.R. Silva, S. Canuto, Phys. Lett. A 101 (1984) 326;
J.R. Silva, S. Canuto, Physica A 106 (1984) 1.
[12] N. Tetradis, C. Wetterich, Nucl. Phys. B 422 (1994) 541.
[13] M. Alford, Phys. Lett. B 336 (1994) 237.
[14] K.I. Aoki, K. Morikawa, W. Souma, J.I. Sumi, H. Terao, Prog. Theor. Phys. 95 (1996) 409;
K.I. Aoki, K. Morikawa, W. Souma, J.I. Sumi, H. Terao, Prog. Theor. Phys. 99 (1998) 451.
[15] B. Boisseau, P. Forgacs, H. Giacomini, J. Phys. A 40 (2007) F215.
[16] P. Amore, F.M. Fernandez, arXiv: 0705.3862.
[17] D.F. Litim, Phys. Rev. D 64 (2001) 105007;
D.F. Litim, Phys. Lett. B 486 (2000) 92.
[18] D.F. Litim, Int. J. Mod. Phys. A 16 (2001) 2081.
[19] D.F. Litim, J. High Energy Phys. 0507 (2005) 005.
[20] T.R. Morris, J. High Energy Phys. 0507 (2005) 027.
[21] C. Bervillier, A. Jttner, D.F. Litim, High-accuracy scaling exponents in the local potential approximation, Nucl.
Phys. B (2007), doi: 10.1016/j.nuclphysb.2007.03.036, hep-th/0701172.
[22] G. Felder, Commun. Math. Phys. 111 (1987) 101.
[23] K.G. Wilson, M.E. Fisher, Phys. Rev. Lett. 28 (1972) 240.
[24] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, The Art of Scientific Computing, Cambridge Univ.
Press, 1992.
[25] R.D. Ball, P.E. Haagensen, J.I. Latorre, E. Moreno, Phys. Lett. B 347 (1995) 80;
J. Comellas, A. Travesset, Nucl. Phys. B 498 (1997) 539;
C. Bervillier, Phys. Lett. A 332 (2004) 93.
[26] D.F. Litim, Nucl. Phys. B 631 (2002) 128.
[27] D.F. Litim, L. Vergara, Phys. Lett. B 581 (2004) 263.
[28] L. Canet, B. Delamotte, D. Mouhanna, J. Vidal, Phys. Rev. B 68 (2003) 064421;
L. Canet, B. Delamotte, D. Mouhanna, J. Vidal, Phys. Rev. D 67 (2003) 065004.
[29] J.P. Blaizot, R. Mendez-Galain, N. Wschebor, Phys. Lett. B 632 (2006) 571;
See also hep-th/0605252;
D. Guerra, R. Mendez-Galain, N. Wschebor, arXiv: 0704.0258.
[30] B. Delamotte, D. Mouhanna, M. Tissier, J. Phys.: Condens. Matter 16 (2004) S883;
B. Delamotte, D. Mouhanna, M. Tissier, Phys. Rev. B 69 (2004) 134413.
[31] T.R. Morris, Phys. Lett. B 334 (1994) 355.
[32] G.E. Andrews, R. Askey, R. Roy, Special Functions, Cambridge Univ. Press, 1999;
G. Gasper, M. Rahman, Basic Hypergeometric Series, Cambridge Univ. Press, 2004.
Towards a universal set of topologically protected gates

for quantum computation with Pfaffian qubits
Lachezar S. Georgiev
Institute for Nuclear Research and Nuclear Energy, Tsarigradsko Chaussee 72, 1784 Sofia, Bulgaria
Received 30 November 2006; received in revised form 10 July 2007; accepted 13 July 2007
Abstract
We review the topological quantum computation scheme of Das Sarma et al. from the perspective of the
conformal field theory for the two-dimensional critical Ising model. This scheme originally used the monodromy properties of the non-Abelian excitations in the Pfaffian quantum Hall state to construct elementary
qubits and execute logical NOT on them. We extend the scheme of Das Sarma et al. by exploiting the explicit
braiding transformations for the Pfaffian wave functions containing 4 and 6 quasiholes to implement, for
the first time in this context, the single-qubit Hadamard and phase gates and the two-qubit Controlled-NOT
gate over Pfaffian qubits in a topologically protected way. In more detail, we explicitly construct the unitary
representations of the braid groups B4 , B6 and B8 and use the elementary braid matrices to implement
one-, two- and three-qubit gates. We also propose to construct a topologically protected Toffoli gate, in
terms of a braid-group based ControlledControlled-Z gate precursor. Finally we discuss some difficulties
arising in the embedding of the Clifford gates and address several important questions about topological
quantum computation in general.
PACS: 11.25.Hf; 71.10.Pm; 73.40.Hm
Keywords: Topological quantum computation; Conformal field theory; Non-Abelian statistics
1. Introduction
In contrast to the three-dimensional world, where we could only find bosons and fermions, the statistics of localized objects in two dimensions turned out to be much richer, and
E-mail address: lgeorg@inrne.bas.bg.

doi:10.1016/j.nuclphysb.2007.07.016
L.S. Georgiev / Nuclear Physics B 789 [FS] (2008) 552590
553
includes fractional or anyonic statistics [1]. In the simplest case of Abelian fractional statistics
the counter-clockwise exchange of two anyons multiplies the many-body state by the statistical
phase exp(i ), where / might be a fractional number. The most important distinction between
anyons on one side and bosons or fermions on the other is that the clockwise and counterclockwise exchanges are significantly different so that the many-body wave functions belong
to the representations spaces not of the permutation group but of the braid group [1,2]. When the
dimension of the braid-group representation is bigger than 1 the corresponding (quasi)particles
are called plektons [35], or non-Abelian anyons, and the exchange of two such anyons results
in a non-trivial statistical matrix acting over a degenerate space of many-body states. As the
matrices representing different exchanges do not commute in general, this kind of quasiparticle
statistics is called non-Abelian.
The most promising two-dimensional system, in which non-Abelian statistics may eventually
be observed seems to be the fractional quantum Hall (FQH) state in the second Landau level
with filling factor = 5/2, which is now routinely observed in ultra high-mobility samples [6,7].
Convincing analytical [8,9] and numerical [10,11] evidence suggest that this state is most likely
in the universality class of the Pfaffian FQH state constructed by Moore and Read [12] using
correlation functions of certain operators in an appropriate (1 + 1)-dimensional conformal field
theory (CFT) [13] including the Ising model [12,14]. Experimental tests of fractional statistics
appear to be much more difficult than those which confirmed the fractional electric charge [15] of
the FQH quasiparticles. However, it turns out that consequences of the presence of non-Abelian
excitations might in fact be easier to observe than the Abelian fractional statistics itself despite
the structural complications. In recent theoretical work interesting proposals for detection of the
non-Abelian statistics of the quasiparticles in the = 5/2 FQH state [1619] and in the = 12/5
FQH state [20,21], which is expected to be in the universality class of the k = 3 parafermion Hall
state [22,23], have been made.
It is quite remarkable that in addition to its fundamental significance, the non-Abelian quantum statistics might become practically important for quantum computation [24]. Although the
ideas behind quantum information processing are simply based on the well-established fundamental postulates of the quantum theory, its exponentially growing computational power could
not have been exploited so far due to the unavoidable effects of quantum noise and decoherence
as a result of interaction of the qubits with their environment. Even the quantum error-correcting
algorithms [24], which allow to use operations containing certain level of errors, could not help
creating a quantum computer with more than a few qubits. In this context recently appeared the
idea of topological quantum computation (TQC) [2527]. Because the interactions leading to
decoherence are presumably local we can try to avoid them by encoding quantum information
non-locally, using some global, e.g., topological characteristics of the system. This topological
protection of qubit operations means that quantum information is inaccessible to local interactions, because they cannot distinguish between the computational basis states and hence cannot
lead to decoherence and noise [25,2729]. That is why topological gates are believed to be exact
operations, which might potentially allow for the topologically protected quantum computation.
The FQH liquid is a perfect candidate for TQC because it possesses a number of topological
properties which are universal, i.e., robust against the variations of the interactions details, and
could be successfully described by rational conformal field (RCFT) theories [3032], which capture the universality classes of the FQH edge excitations (see also [33] and references therein).
The main idea is to use the braiding matrices [3,4,27] corresponding to the exchanges of FQH
quasiparticles, acting over a degenerate set of topological many-body states, to implement arbitrary unitary transformations [2527,34]. Because the single qubit space is two-dimensional, we
554
need degenerate spaces of quasiparticle correlation functions with dimension at least 2. While
the Abelian FQH quasiparticles have degenerate spaces on non-trivial manifolds such as torus
[25,35], such constructions are not appropriate for experimental realization in planar systems,
such as the FQH liquids. Alternatively, by Abelian anyons, one could realize in the plane only
diagonal gates, such as Controlled-phase gates [36] but cannot implement non-diagonal gates
such as the Hadamard gate. In contrast, the non-Abelian FQH quasiparticles by definition have
degenerate spaces even in planar geometry, which explains why the non-Abelian anyons might be
more appropriate for TQC. The only residual source of noise and decoherence is due to thermally
activated quasiparticlequasihole pairs, which might execute unwanted braidings. Fortunately,
these processes are exponentially suppressed at low temperature by the bulk energy gap, which
leads to astronomical precision of quantum information processing.
In a recent paper [28] Das Sarma et al. proposed to use the expected non-Abelian statistics of
the quasiparticles in the Pfaffian FQH state to construct an elementary qubit and execute a logical
NOT gate on it. The NOT-gate construction is very important for quantum computation because
it underlies both the construction of the single-qubit gates and of the Controlled-NOT (CNOT)
gate [24]. However the NOT gate is certainly not sufficient for universal computation, though it
is reversible, because the universal classical gate, NAND/NOT, built from NOT is irreversible
[37], while all quantum gates must be reversible. Therefore, in the TQC scheme of Das Sarma et
al., we need to implement the CNOT gate (or reversible XOR), which plays a central role in the
universal quantum computer [24,37].
The FQH state at = 5/2 has one big advantage with respect to TQCit is the most stable
state, i.e., the one with the highest bulk energy gap, among all FQH states in which non-Abelian
quasiparticle statistics is expected to be realized. Recent measurements of the energy gap [6,7]
suggest that ratio noise/signal is expected to be of the order 1030 or even lower, which is an
unprecedented number in the quantum computation field and might potentially allow for the
construction of a truly scalable quantum computation platform. On the other hand, one serious
drawback is that the quasiparticles braiding matrices cannot be used alone for universal quantum
computation because, as we shall see later, the representation of the braid group turns out to be
finite. This has to be compared with the state at = 12/5, whose braiding matrices are expected
to be universal, but whose energy gap is an order of magnitude lower than the = 5/2 one, which
increases dramatically the noise/signal ratio.
In this paper we extend the TQC scheme of Das Sarma et al. [28], which was originally based
on monodromy transformations of the Pfaffian wave functions, in such a way to construct by
braiding the single-qubit Hadamard and phase gates as well as the two-qubit Controlled-Z and
Controlled-NOT gates. These constructions are naturally topologically protected. In addition we
investigate some possibilities for topologically protected realization of the three-qubit Toffoli
gate.
Summary of results: we review in Section 2 the TQC scheme of Das Sarma et al. and introduce
in Section 2.4 the 4-quasihole wave functions of Ref. [28] which we shall use to derive the
elementary exchange matrices that will represent single-qubit gates. In Section 3 we consider in
more detail the read-out transformation for the qubit of Ref. [28] by using the analytic properties
of the 4-quasiholes states and prove, under certain conditions, the conjecture of Ref. [28] that the
state of the qubit does not change after transferring one quasihole from antidot 1 to antidot 2,
which is crucial for the construction, initialization and manipulation of the Pfaffian qubits. When
these conditions are not satisfied, the TQC scheme of Ref. [28] is going to fail. We prove in
Section 3.3 the orthogonality of the 4-quasiholes states forming the computational basis, which is
very important for their quantum distinguishability. Deriving explicitly in Section 5 the complete
555
set of exchange matrices for the 4-quasiholes states, which has been partially done in Ref. [38]
and completely reproduced in Ref. [39] using the underlying quantum group structure for the
parafermion quantum Hall states, we construct in Section 6 all single-qubit gates, except for the
/8 gate, entirely in terms of quasihole braiding. Then, in Section 7 we propose a natural twoqubit construction in terms of 6-quasiholes states and obtain explicitly the exchange matrices
for these states. In Sections 7.2 and 7.4 we implement the Controlled-Z and Controlled-NOT
gates entirely in terms of 6-quasiholes braidings. To the best of the authors knowledge this is
the first explicit construction of these gates in the Pfaffian state, which is exact and topologically
protected. We also construct in Sections 7.6 and 7.7 the BravyiKitaev two-qubit gate g3 and the
non-demolition charge measurement gate, respectively, in terms of 6-quasiholes braidings. While
the above gates are not sufficient for universal quantum computation they are known to form a
Clifford group [35,40], which plays an extremely important role in error-correcting algorithms
and, in particular could be efficiently used in such applications as quantum teleportation and
super-dense coding [24]. Moreover, if the Clifford group is supplemented by the so-called magic
states or noisy ancillas that could already be used for universal quantum computation [40]. In
addition to the Clifford-group gates, instead of using the /8 gate, we propose in Section 8
to implement in a topologically protected way the Toffoli gate in terms of the CNOT and the
Controlled-S gate, or by a braid-group based ControlledControlled-Z gate precursor, which
would actually form a universal set of topological protected gates realized with Pfaffian qubits.
It appears that there is an additional topological entanglement in the three-qubit systems defined
by 8 Pfaffian quasiholes, which leads to complications in the embedding of the one- and twoqubit gates into systems with three or more qubits. This phenomenon seems to be common for
all topological quantum computation schemes based on the braid matrices of the non-Abelian
FQH anyons.
2. The TQC scheme of Das Sarma et al.
The main idea of Refs. [28,29] is to use the wave functions of the Pfaffian FQH state with 4
quasiholes to form an elementary qubit. Then quantum gates can be executed by braiding some
of the quasiholes (i.e., by counter-clockwise exchanges of quasiholes) leading to unitary transformations in the qubit space. When the positions of the quasiholes are fixed these wave functions
form a 2-dimensional space which could be used as the single-qubit space. In general, the wave
functions containing 2n quasiholes with fixed positions form a linear space with dimension 2n1
[38]. The main reason for the exponential increase of the space dimension is the non-Abelian
statistics of the quasiholes, i.e., when two quasiholes are fused together (taken to the same point
in the coordinate plane) the result contains more than one channels due to the chiral Ising model
fusion rule [12]
= I + .
(1)
The exponential degeneracy of the 2n quasiholes spaces can be alternatively understood by interpreting the Pfaffian state as a p-wave superconductor of composite fermions [8], where the
non-Abelian quasiholes are represented by half-quantum vortices, and their non-Abelian statistics follows from the existence of Majorana zero-modes in the vortex cores obtained as solutions
of the Bogolubovde Gennes equations [8,9,41].
One way to keep the positions of the quasiholes fixed is to introduce antidots [42] (lithographically defined potential hills expelling the FQH fluid and creating a hole or island inside it)
556
Fig. 1. (Color online.) A = 5/2 Hall bar with two antidots, on which the quasiholes (denoted symbolically by small
empty circles) with coordinates 1 and 2 comprising the qubit are localized, and 4 front gates, M, N , P and Q creating
tunneling constrictions. Black arrows depict the edge states, while the green and red arrows denote two alternative
tunneling channels, a direct one and such enclosing the two antidots.
in the FQH liquid and localize the quasiholes there as shown in Fig. 1. The positions of the quasiholes are denoted by a , where a = 1, . . . , 4, and we assume that the quasiholes with coordinates
1 and 2 form our qubit, while 3 and 4 (not shown explicitly in Fig. 1) are used to measure
and manipulate the qubits state [28,43].
2.1. State of the qubit and its initialization
In order to have a real TQC scheme we need to specialize the computational basis {|0, |1}.
To initialize the qubit we put charge 1/2 on antidot 1 [28]. This can be done by adding one quantum h/e of magnetic flux in the vicinity of antidot 1 (e.g., by a solenoid piercing the antidot).
A detailed analysis of the tunneling situation in the stable strong-coupling regime in a Pfaffian
antidot with AharonovBohm flux can be found in Ref. [18]. The FQH liquid containing the antidot responds by localizing a charged excitation on the antidot border carrying one flux quantum.
Because of the fundamental FQH effect relation between the magnetic flux and electric charge1
Qel = , = 1/2, the charge of such an excitation is 1/2. There are only two allowed charge
1/2 excitations of the Pfaffian FQH state which could be localized on the antidot: in the notation
of Refs. [18,33] these are represented by the field operators
i 1 (z)
:e
i 1 (z)
and (z):e
:,
= 1,
Qel = 1/2,
(2)
where (z) is the Ising model Majorana fermion and the vertex exponent of the normalized
boson field (z) represents the u(1) part of the excitation. Now we can envision the following
computational basis
1 As usual we consider only the fractional part of the filling factor = 2 + 1/2 corresponding to the top-most Landau
level.

i 1 (z)
|0 if the charge 1/2 state is :e
:,
i 1 (z)
|1 if the charge 1/2state is (z):e
557
:,
(3)
or, in other words, the state of the qubit is |0 if the Majorana fermion is not occupied and |1
if it is occupied. Note that the states corresponding to the fields in Eq. (2) cannot form coherent
superpositions because they belong to different superselection sectors inside the NeveuSchwarz
sector (when the flux threading antidot 1 is one quantum in the experimental setup), which correspond to their different fermion parity.
The quasiparticle spectrum could be obtained from the chiral partition functions of the corresponding topological sectors, see Refs. [18,33] for more details.
2.2. Measurement of the qubit state
It was one of the bright ideas of Ref. [28] to use the electronic MachZehnder interferometry
[19,44,45] to determine the state of the qubit. More precisely, let us try to measure the diagonal
conductance, xx , which is proportional to the probability for a charged particle to enter through
the lower edge of the Hall bar in Fig. 1 and to exit out of the upper edge. Because of the constrictions created by the front gates M and N , P and Q, there are two alternative channels for a
charged quasiparticle entering from the lower edge to exit from the upper one: one is to tunnel
between the front gates M and N and the other is to tunnel between P and Q. Therefore, to
leading order in the tunneling amplitudes tMN and tP Q , which are assumed to be very small, the
diagonal conductance would be proportional to the interference of the two amplitudes [28]. Consider, for example the interference of the two tunneling processes if the charge 1/2 on antidot 1
does not contain Majorana fermion. Then the amplitude for tunneling between P and Q must be
multiplied by the AharonovBohm phase for the non-Abelian quasihole which when tunneling
between P and Q actually encircles the charge 1/2 state on antidot 1, i.e.,
e2iQel = ei 2 = i,
(4)
because a quasiparticle with charge Qel = 1/4 encircles magnetic flux = 1. Therefore, the
diagonal conductance, which is proportional to the modulus-square of the amplitude, reads
|0
xx
|tMN + itP Q |2 .
(5)
If instead, the state of the qubit is |1, i.e., there is a Majorana fermion on antidot 1, then in
addition to the AharonovBohm phase (4) there would be a minus sign coming from the fact that
the Ising model filed is transported around the Majorana fermion. To see, in the CFT language,
why this minus sign appears, we consider the operator-product expansion
(0)
(z)(0)
z0
2z
so that (z)(0) (z)(0)
when z e2i z,
i.e., when is transported adiabatically around . Thus, the diagonal conductance measurement
in the state |1 gives
|1
xx
|tMN itP Q |2 .
(6)
Note that the two different interference patterns, Eqs. (5) and (6) of the diagonal conductance are
very well distinguished experimentally due to the high visibility of the MachZehnder interferometry [44].
558
2.3. Splitting the 1/2 charge: Finalizing the qubit

Despite that we can efficiently measure the two states in the computational basis this is still not
sufficient for quantum computation. The reason is that we need to form coherent superpositions
of the states |0 and |1 which is not allowed for the charge 1/2 states (3) due to the fermion parity
superselection rule. In order to circumvent this difficulty Das Sarma et al. have made another
interesting proposal [28]: to split the charge 1/2 state into 1/4 1/4 state by transferring one
charge 1/4 from antidot 1 to antidot 2. This is indeed possible, if one applies voltage between
the two antidots, because the most relevant quasiparticle for tunneling through the bulk of the
Pfaffian FQH liquid carries charge 1/4. This quasiparticle is non-Abelian and contains a field
from the Ising model because there is no other charge 1/4 quasiparticle and the non-Abelian one
has the minimal CFT dimension [12,33,38]. Now the state on antidots 1 and 2 is equivalent to
(1 ) (2 ) which together with the quasiholes at 3 and 4 correspond to a 4-quasihole wave
function, which belongs to a two-dimensional space as discussed in the beginning of Section 2.
While the charge 1/2 states (3) cannot form coherent superpositions, because they belong to
different superselection sectors, for the 4 quasiholes configurations we can consider the superposition of the states obtained by splitting of the vacuum in two different ways, namely
I I I ( ) ( ),
I ( ) ( ),
(7)
because the two 4-quasihole states now belong to the same superselection sectors.
The original idea of Ref. [28] was that the state of the qubit does not change during the
splitting procedure because the pairs of quasiholes in the Pfaffian state share a pair of Majorana fermions zero modes, (0 , 0 ), whose combined (left- plus right-moving) fermion parity is
supposed to be conserved. In Section 3 we shall explicitly derive the readout results for the
4-quasiholes wave functions and will find conditions under which this is in agreement with
Ref. [28].
2.4. Four-quasiholes wave functions
As we shall see in Section 3, the topological phase, which determines the conductance interference pattern, can be obtained from the monodromy matrices appearing in the adiabatic
transport of some quasiholes around others, so that we need the wave functions explicitly. The
wave function for even number N of holes (or electrons) at positions z1 , . . . , zN containing 4
quasiholes at positions 1 , . . . , 4 , can be realized as a correlation function
4qh 1 , 2 , 3 , 4 ; {zi } = qh (1 )qh (2 )qh (3 )qh (4 )
N

hole (zi )
i=1
of the field operators corresponding to creation of holes and quasiholes

hole (z) = (z):ei
2(z)
: and qh () = ():e
()
2 2
:,
(8)
respectively, where () is the chiral spin field in the Ising model of dimension 1/16 and (z)
is the right-moving Majorana fermion in the chiral Ising model. It can be expressed as
559
4qh (1 , 2 , 3 , 4 ; z1 , . . . , zN )

N

= (1 ) (2 ) (3 ) (4 )
(zj )
j =1

1a<b4
1/8
ab
N

(zi 1 )(zi 2 )(zi 3 )(zi 4 )
2
zij
,
(9)
1i<j N
i=1
where the average sign now represents the vacuum expectation value in the chiral Ising
model
and the last three product factors in Eq. (9) come
from the u(1) components of the holes
:exp(i 2(z)): and of the quasiholes :exp(i(z)/2 2 ):. We used here the notation ab a
b for a
= b and zij = zi zj for i
= j .
One important detail is that the chiral field () does not have a definite chiral fermion parity
because of the non-Abelian fusion rule (1) which mixes states with different fermion parities.
Nevertheless it would be convenient to introduce two chiral fields of dimension = 1/16
and opposite parities [33]
F ()F = (),
(10)
in terms of which the non-Abelian fusion rule (1) necessarily splits into two Abelian channels
= I,
= .
(11)
The filed entering Eq. (9) is then identified with

+ () + ()
(12)
.
2
In order to obtain the 4-qh wave function (9) we carefully repeat the arguments of Ref. [38] in
which the chiral many-body wave functions of the Pfaffian state are obtained by bosonization
techniques from the c = 1 complex Ising model. The final result is
() =
(0)
(1)
4qh (1 , 2 , 3 , 4 ; z1 , . . . , zN ) = 4qh + 4qh ,
(13)
(0)
(1)
|0, 4qh
|1 would be our compuwhere we have used the notation of Refs. [28,38] (4qh
tational basis in the 4-qh wave functions space)
(13 24 )1/4
(0,1)
4qh =
((13)(24) x(14)(23) )
1 x
with x being a CFT invariant crossratio [13]
14 23
x=
and (ab = a b ),
13 24

(zi a )(zi b )(zj c )(zj d ) + (i j )
(ab)(cd) = Pf
zi zj
(14)
(zi zj )2 ,
1i<j N
(15)
where {a, b, c, d} is a permutation of {1, 2, 3, 4} satisfying a < b and c < d. The Pfaffian of an
anti-symmetric matrix Mij of even dimension N is defined as
N/2

1
sign(
)
M (2k1) (2k) .
Pf(Mij ) = N/2
2 ((N/2)!)2
SN
k=1
560
It is worth stressing that the space of the 4-quasihole wave functions for fixed positions of the
quasiholes is two-dimensional. The second independent wave function could be obtained from
the first one, Eq. (13), by transporting 3 around 4 (i.e., 34 e2i 34 ) which transforms
(0)
(0)
(1)
(1)
4qh 4qh and 4qh 4qh so that
(0)
(1)
4qh (1 , 2 , 3 , 4 ; z1 , . . . , zN ) = 4qh 4qh .
(16)
It appears, however, that in some situations, such as the qubit initialization of Ref. [28], the 4-qh
(0)
(1)
or 4qh
.
wave function may be driven directly in the state 4qh
3. Read-out in the TQC scheme of Das Sarma et al.: The measurement of the 4-quasiholes
qubit state
As we have seen in Section 2.2 the read-out of the qubit is performed by interference measurement of the diagonal conductance, for which we need the monodromies of the corresponding
wave functions. In this section we shall compute the explicit monodromies of the 4-qh wave functions (14) as well as for (13) and (16).
Assuming that our qubit is formed by the quasiholes with coordinates 1 and 2 we can
interpret the quasihole with coordinate 3 tunneling either through M and N or through P and
Q as generating the interference pattern in the longitudinal conductance, like in the non-Abelian
MachZehnder interferometer [45]. In more detail, if the third quasihole tunnels through M
and N , this could be interpreted as a clockwise braid of this quasihole with the qubit, while if
it tunnels through P and Q this gives rise to a counter-clockwise braid (denoted below as R;
cf. Ref. [45]) so that the quantum amplitude for these two processes is (we have absorbed all
dynamical phase factors for the two paths in the corresponding amplitudes tMN and tP Q ) |A =
tMN R 1 |(1 , 2 ), 3 , 4 + tP Q R|(1 , 2 ), 3 , 4 and the longitudinal conductance would be

tP Q |R 2 | ,
xx A|A = |tMN |2 + |tP Q |2 + 2 Re tMN
where | = |(1 , 2 ), 3 , 4 is the 4-qh state in which the qubit is formed by the quasiholes
with positions 1 and 2 , and we have used the unitarity of the braid operator R. The monodromy
operator R 2 is actually the operator which takes the quasihole with coordinate 3 around those
with 1 and 2 in counter-clockwise direction. Note that if we fuse the quasiholes forming the
qubit, i.e., 1 2 , and the state is I then the matrix element is +i, while if they fuse to the
matrix element is i, which reproduce the interference results in Section 2.2. It is worth stressing
that there is a remarkable relation between the expectation value of the monodromy operator R 2 ,
corresponding to the adiabatic transport of particle with label a around particle with label b, and
the modular S matrix [13] associated with any rational CFT [20]:
|R 2 | =
Sab S00
,
S0a S0b
where the label 0 corresponds to the vacuum. Working with modular S matrices is very convenient because they are explicitly known for almost all rational CFT related to the FQH effect.
Thus we see that the read-out or measurement procedure for the 4-qh wave functions consists in taking the quasihole with coordinate 3 around the qubit [28,43], i.e., around the two
quasiholes with coordinates 1 and 2 .
It is worth stressing that the 4-qh wave function basis (14) is very convenient from the point of
view of the adiabatic transport because the functions (ab)(cd) have no Berry phases [38] so that
561
Fig. 2. The trivial monodromy of the 4-quasihole wave function, when one quasihole is transported around all others,
allows to compute the read-out as taking 3 around 4 .
the transport effects could be explicitly determined from the monodromies of the multivalued
function

f (z) = 1 z, z C,
entering the denominator in Eq. (14). Notice, however, that taking 3 around 1 and 2 in the
functions (19) for z = x with x defined in Eq. (15), which are simply proportional to the 4-point
functions of the chiral spin field in the Ising model, is equivalent to the adiabatic transport of
3 around 4 alone. This is because the monodromy for the transport of 3 around 1 , 2 and
3 is trivial, as shown in Fig. 2, since the contour on the left can be contracted to a point (at
infinity) without passing through any other singularity. Therefore it would be simpler to compute
the read-out transformation of the functions (19) below by
2
:
R34

34 34
= lim e2it 34 ,
t1
t [0, 1).
The transformation (17) obviously preserves the absolute value of the crossratio
12 34
= 1 x x = 1 x,
x
13 24
(17)
(18)
= , i.e., the contour for transportation

where x is defined in Eq. (15), because all other ab
ab
of x,
corresponding to Eq. (17), is a circle with center at x = 0 and radius |x|.
Therefore the
result crucially depends on whether |x|
is bigger or smaller than 1. Thus we need to consider the
behavior of the functions

= 1 1 x
f (x)
(19)
under the transformation (17). Yet, it is more convenient to first analyze Eq. (19) as a function
of z and then just outline what happens when we change z = 1 x.
The multivalued function (19) has two separate branching points: one is z = 0 which, if encircled by the continuation contour, would change the sign of the inner root, as can be seen from
Eq. (A.2) in Appendix A, i.e.,

1 z 1 z, for z e2i z and |z| < 1.

However, the second function, corresponding to the minus sign under the square root in Eq. (19),
has one more branching point at z = 1 which, if encircled in the process of the analytic continuation changes the sign of the outer root, i.e.,

1 z 1 z, for (z 1) e2i (z 1) and |z 1| < 1.
562
Fig. 3. The contour Cx<

used for the analytic continuation of Eq. (19) when |x|
< 1. As it encloses only the branching
point at z = 1, going along this contour only changes the outer-root sign of f (z).
This can be easily seen from the Laurent-mode expansion for |z 1| < 1, by looking at Eq. (A.4)
in Appendix A.
Now consider what happens when we change z = 1 x.
When x e2it x,
and t goes from 0
to 1, this transports x along a circle of radius |x|
and center at x = 0. Then x is transported
along the same circle, though with a phase shift of . Thus z = 1 x is transported along a circle
of the same radius |x|,
however translated to be centered at z = 1 as shown in Figs. 3 and 4.
Therefore we need to consider both cases, |x|
< 1 and |x|
> 1, separately in more detail. Of
course, one may consider other contours which are homotopic to that of Eq. (17) but do not
preserve the absolute value of x.
In that case the results would be the same as those obtained
from Eq. (17), only the conditions on |x|
must be replaced by the homotopic condition whether
these contours encircle both branching points z = 0 and z = 1 or only one of them.
3.1. The read-out for |x|
< 1: Take 3 around 4
According to our analysis, when |x|
< 1, the read-out transformation (17) corresponds to
transporting the crossratio (18) along a contour which encircles only the branching point at z = 1,

2it
Cx<
(20)
x | 0 t 1, |x|
<1 ,
= z=1e
as shown in Fig. 3. We have introduced two branch-cuts for the function (19) like in Fig. 3.
Because the 4-quasihole wave functions (13)(24) and (14)(23) have no branch-cuts in ab , they
acquire no phase under the transformation (17), and the only phases come from the prefactor in
Eq. (14) containing ab and x.
Since |x|
< 1 and the transformation (17) transports x along the contour (20), which is shown
on Fig. 3, this transformation only changes the outer-roots signs, i.e.,

1 + 1 x
1 + 1 x
1 0
2

R34
=
.
0 1
1 1 x
1 1 x
(0)
(1)
In order to find the action of the transformation (17) in the basis {4qh , 4qh } we only have to add
1/4
the phase coming from 13 in Eq. (14) which is ei/2 . Notice that we use taking 3 around 4
563
Fig. 4. The contour Cx>

used for the analytic continuation of Eq. (19) when |x|
> 1. Because it encloses both branching
point at z = 1 and z = 0 going along this contour changes the outer-root sign whenever the inner one is and flips the
inner-root sign of f (z).
as the read-out prescription only for the functions (19) because these are the functions for which
the total monodromy is trivial. For all other functions, including fractional powers of ab , we still
use as the read-out taking 3 around 1 and 2 . Thus we obtain the read-out transformation as
(0)

(0)
4qh
4qh
12 34
i 0
(|x|<1)

=
for
Uread-out
(21)
< 1.
(1)
(1)
0 i
13 24
4qh
4qh
Notice that this i, on the diagonal of the matrix in Eq. (21), is exactly the topological phase
that appears for the quasiparticles traveling along the contour passing through the front contacts P and Q, see Fig. 1, in the diagonal conductance measurement so that
|x|<1
|tMN itP Q |2 ,
xx
with + for |0 and for |1.
(22)
3.2. The case |x|

>1
Alternatively, if |x|
> 1, then the read-out transformation (17) leads to transporting x along
the following contour

2it
x | 0 t 1, |x|
>1 ,
Cx>
(23)
= z=1e
which encircles both branching points at z = 1 and z = 0 as shown in Fig. 4. Therefore, the readout transformations (17) would change the sign of the outer root whenever the inner sign is
as well as would flip the inner-roots signs, i.e.,

1 + 1 x
1 + 1 x
0 1
2

R34
=
.
1 0
1 1 x
1 1 x
564

1/4
Adding again the Abelian phase ei/2 , coming from 13 in Eq. (14) we obtain the read-out
transformation as
(0)

(0)
4qh
4qh
12 34
0 i
(|x|>1)

Uread-out
(24)
=
for
> 1.
(1)
(1)
i 0
13 24
4qh
4qh
Now the diagonal conductance measurement might be different from Eq. (22), as it crucially de(0,1)
pends on the overlap between the states (14). If the two states 4qh are orthogonal, which is a
(0,1)
(0,1)
2 |
fundamental requirement in quantum theory, then the monodromy average 4qh |R34
4qh ,
determining the interference pattern, would vanish and the diagonal conductance could not distinguish between the two states in the computational basis. Thus we see that the MachZehnder
interference measurement could only work for |12 34 | < |13 24 |.
Using the CFT invariance of x,
one can prove that the absolute value of the crossratio x
depends on the absolute values of the quasiholes positions, i.e.,
|x|
<1
|3 | < |2 |,
while |x|
>1
|3 | > |2 |.
Remark 1. The main result in this section has very important implications for the experimental realization of the topological quantum computer in FQH systems at = 5/2. The read-out
procedure crucially depends on the absolute value of the crossratio (18). When |x|
< 1 the TQC
scheme of Ref. [28] is going to work as originally proposed, while if |x|
> 1 this scheme would
fail. As the absolute value of x depends on the quasiholes positions, this certainly gives some
hints on how the antidots should be arranged in order for the read-out procedure to work as in
Ref. [28] and, as we shall see later, this is also related to whether the NOT gate could be executed
in the Pfaffian qubit or not.
(0,1)
3.3. Orthogonality of the 4-quasiholes wave function 4qh
One of the fundamental requirement in quantum theory is that the states |0, |1 are not only
linearly independent but in fact orthogonal. Otherwise there exists no measurement that can
reliably distinguish these states [24]. Here we will demonstrate that the 4-quasiholes states (14)
are indeed orthogonal, at least for |x|
< 1. This seems natural because, as we already know
the MachZehnder interferometer can distinguish the states (14), hence these states should be
orthogonal. However, it is still worth verifying this directly if possible.
(0)
(0)
(1)
The point is that for |x|
< 1 we have, according to Eq. (21), U 4qh
= i4qh and U 4qh =
(1)
(0)
(0)
(1)
(1)
i4qh , hence, U 4qh = i4qh and U 4qh = i4qh , where U Uread-out . In order to find
(1)
(0)
the overlap (4qh , 4qh ) we first plug inside it the operator U and use the above relations, i.e.,
on one side we have
(1)
(1) (0)
(0)
4qh , U 4qh
(25)
= i 4qh , 4qh
while, on the other side, using the definition of the Hermitean conjugation and the anti-linearity
of the inner product with respect to its first argument, it is equal to
(1) (0) (1) (0)
(1) (0)
U 4qh , 4qh = i4qh , 4qh = i 4qh
(26)
, 4qh .
Obviously the right-hand sides of Eqs. (25) and (26) could only be equal if
(1) (0)
4qh , 4qh = 0 for |x|
< 1.
565
4. The NOT gate of Das Sarma et al.

The original idea of Refs. [28,29] behind the construction of the logical NOT gate in the setup
shown in Fig. 1, is that when the quasiholes with coordinates 1 and 2 , localized on antidots 1
and 2, form the qubit [43] another quasihole with coordinate 3 could tunnel between the front
gates A and B (through an additional antidot located between A and B in order to guarantee a
single quasihole tunneling) in such a way to flip the state of the qubit. The point is that this third
quasihole actually encircles the quasihole localized on antidot 1 but not the second quasihole
localized on antidot 2. More generally, while the read-out of the qubit is performed by taking 3
around 1 and 2 , the NOT gate could be executed by taking 3 around 1 or 2 but not both of
them.
To understand this in more detail let us consider the simplest situation when the quasiholes
are ordered |1 | > |2 | > |3 | > |4 | and the quasihole at 3 traverses a closed loop around the
quasihole at 2 , i.e.,
NOT:
23 lim e2it 23 ,
t1
where t [0, 1),
(27)
with all other quasiholes coordinates remaining unchanged. Provided that the three quasiholes at
1 , 2 and 3 are kept well separated, after the tunneling of the third quasihole the 4-qh wave
functions (14) are adiabatically transformed into new wave functions which we shall find now.
First the transformation (27) transforms the crossratio (15) according to x e2i x. Next we
note that the Pfaffian wave functions (ab)(cd) are single-valued, so that they remain unchanged
under the NOT transformation (27). Finally, the square root of x changes sign under the NOT
transformation (27), i.e.,

x e2i x = ei x.

The transformation of 1 x is
and depends on the absolute value of the crossmoresubtle
ratio x. When
|x|
<
1
we
have
1 x under the transformation (27), while
1

if |x| > 1, 1 x 1 x, because the transformation contour now encircles both

branching points. Thus we find that the two 4-qh wave functions (14) transform under the NOT
transformation (27) as follows
(0)
(0)
4qh
4qh
0 1
Das Sarma
UNOT
(28)
=
for |x| < 1, while
(1)
(1)
1 0
4qh
4qh
(0)
(0)
4qh
0 1
4qh
Das Sarma
UNOT
(29)
=
for |x| > 1.
(1)
(1)
1
0
4qh
4qh
(0)
(1)
Thus we conclude that the transformation (27) indeed maps 4qh
4qh
.
5. Exchange matrices and monodromy group representation of the Ising 4-point functions
The four-point correlation functions of the chiral spin field (12) in the Ising model can be
shown to be [46]

F (1 , 2 , 3 , 4 ) (1 ) (2 ) (3 ) (4 ) = F+ + F ,
1/8

14 23
1
13 24
where F (1 , 2 , 3 , 4 ) =
(30)
1
,
13 24
2 12 14 23 34
566
are the chiral conformal blocks, which could be expressed in terms of the fields with definite
fermion parity as F = + (1 ) (2 )+ (3 ) (4 ).
The counter-clockwise exchange of the quasiparticles with coordinates 1 and 2 can be preformed by analytic continuation along the contour defined by
1 2
1 2
1 + 2
1 + 2
(31)
+ eit
, 2 =
eit
, 0 t 1.
2
2
2
2
Executing this transformation and taking the limit t 1 we obtain for the 4-pt functions (note
= ei )
that 12
12
i

13 24
e 13 24 1/8 14 23 1/4
1+
= ei 8 F (a ).
R12 F =
12 14 23 34
13 24
14 23
1 =
Therefore (fixing the signs of the square roots by the requirement that the R-matrices must satisfy
the YangBaxter equations (33)) we get a diagonal matrix

F+
F+
1 0
R12
= ei 8
.
0 i
F
F
Precisely in the same way we can compute the exchange matrix R34 which takes precisely the
same form in this basis. In order to compute the exchange matrix R23 we apply the coordinate
transformation (31), after renaming (1 , 2 ) (2 , 3 ), and use the identity

1 + x + i 1 x = 2i
1 x i x
for = 1. We obtain
1/8

ei 8
ei 8 i
13 24
1 x i x = e 4 (F+ iF ).
R23 F =
2 12 14 23 34
2
Thus, we can summarize our results for the elementary exchange matrices Ra,a+1 in the basis
{F+ , F }

ei 8
1 0
1 i
,
R23 =
.
R12 = R34 = ei 8
(32)
0 i
2 i 1
All other exchanges can be obtained from the elementary ones, e.g.,

ei 8
1 1
1
, etc.
R13 = R12 R23 R12 =
2 1 1
(0,1)
5.1. Exchange matrices for the 4-quasiholes wave functions 4qh
The 4-quasiholes wave functions (14) are built up from the 4-point functions of the Ising
model and the functions (ab)(cd) which are single-valued in the positions of the quasiholes.
Now a natural question arises: is the braid-group representation for the 4-point functions (19)
extended to a braid-group representation over the 4-quasihole functions? The answer is yes. The
point is that when a quasihole traverses a closed loop, the 4-quasihole functions should acquire
an additional phase proportional to the number of electrons inside the loop (or, to the area of the
loop). This might lead to a projective representation of the braid group as looks to be the case in
the p-wave composite-fermion superconductor approach, where the quasihole is identified with
567
a half-quantum vortex [9]. However, as can be seen in our approach directly from the 4-quasihole
functions (14), each electron inside the loop contributes 2 to this phase because the quasiholes
are by definition local with the electrons.2 Thus the phase is insensitive to the number of electrons
inside the loop, it only counts the number of quasiholes. Nevertheless it turns out that braid-group
(or mapping class group) [2] representations in terms of CFT correlation functions are generically
projective. The point is that the coordinates of the many-body wave functions, which in the
CFT approach [33] are chiral correlation functions defined on the unit circle, could be naturally
extended by analytic continuation to the vicinity of the unit disk. Then by CFT transformation
these functions could be extended to the entire compactified complex plane, which is isomorphic
to the two-dimensional sphere. Now, besides the Artin relations [2]
Bi Bj = Bj Bi ,
for |i j | 2,
Bi Bi+1 Bi = Bi+1 Bi Bi+1 ,
where Bi = Ri,i+1 Bn ,
(33)
for the generators Bi (i = 1, . . . , n 1) of Bn , the representation of the braid group on the sphere
should satisfy one more relation [2]
2
Bn2 B2 B1 = I.
B1 B2 Bn2 Bn1
As can be seen by direct computation the above relation is satisfied by the elementary exchange
matrices of the chiral CFT correlators only up to phase so that the braid-group representation is
projective.
In the rest of this subsection we will show that from the exchange matrices for the functions
(0)
(1)
F we could obtain the corresponding matrices in the basis {4qh , 4qh } in the form [47]

ei 4
1 i
1 0
(4)
(4)
(4)
R12
(34)
,
= R34
=
,
R23
=
0 i
2 i 1
where the superscript (4) is to remind us that these matrices are computed in the basis of the 4quasiholes states (14). Consider, for instance, the transformation (31) acting on the 4-quasiholes
wave functions (14). It changes x 1/x and (13)(24) (14)(23) so that

(23 14 )1/4
1
(4) (0,1)
R12 4qh =
(14)(23) (13)(24) .

x
1
1 x

Expressing 23 14 = 13 24 x and taking out 1/x from the denominator we obtain the
(4)
exchange matrix R12 in the basis (14) as in Eq. (34). Notice that there is no more ei/8 in
this matrix. At this point the YangBaxter equations (33) imply that if there is a braid-group
representation over the 4-quasihole wave functions (14) they must be obtained from Eq. (32)
(4)
by multiplying all exchange matrices with ei/8 . Because R12 is diagonal it could be directly
obtained by first fusing 1 2 and then interpreting the exchange as 12 ei 12 , which
gives the same result.
(4)
(4)
Next, instead of computing R23 directly it is more convenient to compute R13 , following
Ref. [38], and then use the identity
(4)
(4) (4) (4) 1
R23 = R12 R13 R12
(35)
.
2 What could really change the phase is not the entire electron but the neutral Majorana fermion, which is non-local
with the quasihole.
568
To this end we first apply the coordinate transformation

1 = 3 ,
2 = 2 ,
3 = 1 ,
4 = 4 ,

such that 13
= ei 13 ,
x x = 1 x and (13)(24) (13)(24) , while (14)(23) (12)(34) . Then, using the Nayak
Wilczek identity [38] (note the misprints in Eq. (3.8) there)
(1 x)(12)(34) = (13)(24) x(14)(23) ,

as well as 1 + 1 x 1 1 x = 2 1 x,
(4)
(36)
(4)
it is not difficult to derive R13 , hence obtain R23 by Eq. (35), i.e.,

(4)
R13
(0)
4qh
(1)
4qh
ei 4
=
2
1 1
1 1
(0)
4qh
(1)
4qh
The sign ambiguity in front of the square root in the right-hand side of Eq. (36) comes from
taking a square root and is not directly linked to the sign under the square root on the right. In
Das Sarma for |x| < 1, that
order to fix this sign we use Eq. (35) and the fact that the NOT gate UNOT
we obtained in Eq. (28) directly in terms of the 4-quasiholes wave functions monodromies in
(4)
Section 4, is actually the square of the exchange matrix R23
, i.e.,
(4) 2 (0,1)
R23 4qh
(1,0)
= 4qh
(4) 2
R23

=

0 1
.
1 0
Remark 2. The relevance of the Pfaffian qubit for quantum computation can be emphasized once
(4)
again. Because the monodromy matrix (R23 )2 coincides with the NOT gate in the 4-quasiholes
(4)
states basis, the matrix R23 should be identified with NOT, which cannot be implemented in
classical information theory [48]. As we shall see later this operation is crucial for the construction of the Hadamard gate, which is one of the most important single-qubit quantum gate.
Remark 3. The derivation of the elementary exchange matrices (34) generating the entire twodimensional representation of the braid group B4 , over the 4-quasiholes Pfaffian wave functions,
(4)
has been exfollows the lines of Ref. [38] where the first row of the exchange matrix R13
plicitly computed. These matrices can be obtained from the general representation of the braid
group B4 for the Pfaffian state, as derived in Ref. [39], using the quantum group structure of
the Zk parafermion Hall states, see Eqs. (138) and (140) there. Nevertheless, the direct and selfcontained derivation of the exchange matrices, given above, from the analytic properties of the
Pfaffian wave functions has certain advantages and provides an independent check of the results.
The monodromy transformations, corresponding to the elementary exchanges, i.e., the squares
(4)
of Ra,a+1
representing a complete counter-clockwise cycle of the particle with label a + 1 around
that with label a,

(4) 2 (4) 2
(4) 2
1 0
0 1
R12 = R34 =
,
R23 =
,
0 1
1 0
569
form a subgroup of the braid group called the monodromy group. The monodromy group representation is thus generated by the Pauli matrices 1 and 3 , or, alternatively, by the two elements3
S = i2 , R = 1 , such that S 4 = R 2 = I and R 1 SR = S 1 . Therefore the image of the monodromy group is isomorphic to the finite non-Abelian group D4 [49,50], known as the symmetry
group of the square, which has 8 elements typically given in the two-dimensional representation
as
D4 {I2 , 1 , i2 , 3 },
where i are the three Pauli matrices. The monodromy group Mn is in general a normal subgroup
of the braid group Bn and the factor-group
Bn /Mn Sn
is isomorphic to the permutation group Sn . Therefore, the fact that the monodromy group representation M4 = D4 for the 4-pt functions is finite implies that the braid group representation
in this case is a finite group whose order could be shown to be | Image(B4 )| = 96. The order of the representation of the braid group B6 is | Image(B6 )| = 46080 = 26 6!, while that of
the monodromy subgroup is | Image(M6 )| = 32. Similarly, the order of the representation of
the braid group B8 is | Image(B8 )| = 5160960 = 27 8!, and that of its monodromy subgroup
is | Image(M8 )| = 128. These numbers have been obtained by direct enumeration of the distinct matrices, produced by multiplying the elementary braid matrices, in the corresponding
braid-group representation using the Diminos algorithm [51]. In general the image of the representation of the braid group B2n , for n 3, over the 2n-point functions of the Ising model
is [52]

2n1 (2n)! for n = even,

Image(B2n ) = 2
for n = odd.
22n (2n)!
While this may look very nice as a mathematical fact it is fairly disappointing from the perspective of topological quantum computation. The point is that our intention in TQC is to implement
quantum gates by simply exchanging quasiparticles positions and the finite braid-group representation implies that we could generate only finite number of gates with the Pfaffian qubit.
Therefore it cannot be used for universal TQC where we would like to efficiently implement
any unitary matrix (with a given precision). However, the set of quantum gates that could be
realized by braiding of Pfaffian quasiparticles is known to be a Clifford group, which plays a
central role in quantum error correction codes, so their topologically protected construction is
fairly important.
6. Single-qubit gates constructed from elementary exchange matrices
In this section we shall implement by quasihole braiding the following single-qubit gates: the
Hadamard gate H , Pauli gates X, Y , Z, and the phase gate S [24]. In Section 7.4 we shall also
construct entirely in terms of quasihole braidings the Controlled-NOT gate. While not sufficient
for universal QC, these gates are known to form a Clifford group [40] and could in principle
be used for universal quantum computation provided that we can create the so-called magic
3 Note that the second Pauli matrix here appears in this context naturally multiplied by i and this is in accord with the
standard quantum computation conventions [24].
570
Fig. 5. Braiding diagram for the Hadamard gate (37) and its symbol (on the right) in standard quantum-computation
notation.
Fig. 6. Braiding diagram for the Pauli X gate and its quantum-computation symbol.
states [40]. The only single-qubit gate that we cannot construct directly by braiding is the /8
gate T [24]. However, instead of the /8 gate one we could use the three-qubit Toffoli gate [24].
The Hadamard gate H is of central importance for any QC scheme. It is worth stressing that
the Hadamard gate is the only gate which must be added to a universal classical computer (based
on the Toffoli gate) in order to make it a universal quantum computer [53]. In the TQC with
Pfaffian qubits it can be used to create special superpositions, called the Bell states (or EPR
states) [24], that can be constructed in no other way. In addition H is one of the building blocks
of the quantum Fourier transform [24]). The Hadamard gate in the Pfaffian TQC scheme can be
expressed in terms of three elementary braidings of the 4-quasiholes states (14), namely,4
ei 4
2
H R12
R13 = R12 R23 R12 =
2
1
1

1
.
1
(37)
It would be convenient to represent these braidings in braid diagrams, such as Fig. 5. The first
and the third braids in Fig. 5 are in clockwise direction and correspond to the inverse exchanges
1
, while the second exchange is in counter-clockwise direction and corresponds to R23 .
R12
The Pauli X gate, known also as the NOT gate, which was first implemented for the Pfaffian
qubit in Ref. [28], could be executed by two elementary exchanges as

2
X R23
=
0 1
1 0
and the corresponding braid diagram is shown in Fig. 6. Similarly, the Y gate, which is usually
defined in quantum computation literature without the imaginary unit, is realized by 4 elementary
exchanges as follows
Y
1 2
R12
R23 R12

=
0 1
1 0
4 The results in this section have been originally announced in Ref. [47]; here we give a detailed derivation.
571
Fig. 7. Braiding diagram for the Pauli Y gate.
Fig. 8. Braiding diagram for the Pauli Z gate.
Fig. 9. Braiding diagram for the phase gate S.
and the corresponding braid diagram is shown in Fig. 7. The Pauli Z gate [24] can be realized in
two different ways by two elementary exchanges as

1 0
2
2
Z R12
= R34
=
0 1
and the braid diagram for the first of them is shown in Fig. 8. Finally, the phase gate S, which
can also be realized in two different ways by a single elementary exchange,

1 0
S R12 = R34 =
0 i
and the first of them, S = R12 , is shown graphically in Fig. 9. Notice that S 2 = Z as it should be.
The only single-qubit gate which cannot be implemented directly in terms of quasiholes braid(4)
ing is the /8-gate T = diag(1, ei/4 ) [24] because det T = ei/4 , while det(Ra,a+1 ) = i for
all a. Instead of T we shall propose to construct the Toffoli gate.
7. Two-qubits construction and two-qubit gates
In order to realize two qubits, which belong to C2 , we need at least 4-dimensional space.
Recalling that the dimension of the excited Pfaffian states with 2n quasiholes at fixed positions
[38] is dim H2n = 2n1 we consider the 6-quasihole states. Before we explain how to construct
the two-qubit states let us recall that the single qubit states can be written as
|0 + + + + ,
|1 + + ,
where we take the convention that the first two fields determine the state of the qubit while
the last two guarantee the preservation of the fermion parity, so that basically |0 + + , while
572
Fig. 10. Two qubits, realized by 6 quasiholes spanning the needed 4-dimensional space C2 . The state of qubit 1 is
determined by the quasiholes with positions (1 , 2 ), while the state of qubit 2 by the quasiholes with positions (5 , 6 )
and that is why these two groups were shaded. The state of the quasiholes with coordinates (3 , 4 ) depends on both
qubits.
|1 + , which is in agreement with our definition of the qubit because of the fusion rules
+ + I and + .
The two-qubit basis is defined here by the convention that the first two quasiholes form the
first qubit while the last two quasiholes form the second qubit as shown in Fig. 10.
|00 + + + + + + ,
|01 + + + + ,
|10 + + + + ,
|11 + + + + .
(38)
This is convenient since if we fuse the first two quasiholes this would project to the second qubit,
while if we fuse the last two quasiholes this would project to the first qubit, i.e.,
| |,
1 2
| |.
5 6
(39)
Then the third and the fourth quasiholes are fixed by the conservation of the fermion parity, i.e.,
if ei is the parity of ei , consider the correlation function with F plugged in the middle. This
gives
e1 e2 e3 F e4 e5 e6 = e1 e2 e3 e1 e2 e3 e4 e5 e6
if we move F to the left or e4 e5 e6 e1 e2 e3 e4 e5 e6 if we move it to the right. Therefore we
obtain the fermion parity rule
e1 e2 e3 = e 4 e5 e6
e 3 e4 = e 1 e2 e5 e6 ,
where the last equality follows from ei2 = 1. Thus we have only 4 independent states in the space
of 6-quasiholes with fixed positions, which correspond to Eq. (38).
7.1. Exchange matrices for 6-quasiholes
The braid group representation over the 6-pt functions is generated by the elementary ex(6)
(6)
(6)
(6)
(6)
changes R12 , R23 , R34 , R45 and R56 . We shall construct them explicitly by using the operatorproduct expansions of the Ising model and the projections to the single-qubit states along the lines
of Ref. [47].
The construction (38) of the two-qubit states allows to derive the 6-quasiholes exchange ma(6)
trices in terms of those for the 4 quasiholes. In order to obtain the exchange R12 we may first
fuse 5 6 , because the state of the first qubit is independent of 5 and 6 , and use the projections (39). In more detail, we have
|00 + + + + ,
|01 + + + ,
|10 + + ,
|11 + + + .
5 6
5 6
573
5 6
5 6
(40)
(4)
Now the exchange 1 2 is represented by the action of R12 , which we take from Eq. (34),
i.e.,
|00 + + + + ,
|01 + + + ,
|10 i+ + ,
|11 i+ + +
1 2
1 2
1 2
1 2
and restoring back the second qubit we obtain in the basis (38)
1 0 0 0
0 1 0 0
(6)
(4)
R12 =
= R12 I2 .
0 0 i 0
0 0 0 i
(41)
(6)
Now let us compute the exchange matrix R23 . Again we can fuse 5 6 and use Eq. (40). The
(4)
exchange 2 3 is represented by the action of R23 , from Eq. (34), i.e.,
ei 4
ei 4
|00 + + + + i + + ,
2 3
2
2
ei 4
ei 4
|01 + + + i + + + ,
2 3
2
2
i 4
i 4
e
e
|10 + + i + + + + ,
2 3
2
2
ei 4
ei 4
|11 + + + i + + +
2 3
2
2
and restoring back again the second qubit according to Eq. (40) we obtain in the basis (38)
1
0 i 0
i 4
e 0
1
0 i
(6)
(4)
R23
(42)
=
= R23 I2 .
1
0
2 i 0
0 i 0
1
Precisely in the same way, by fusing first 1 2 , we can obtain the 6-quasiholes exchange
matrices
1 i 0
0
i 4
e i 1
0
0
(6)
(4)
R45 =
(43)
= I2 R23 , and
0
0
1
i
2
0
0 i 1
1 0 0 0
0 i 0 0
(6)
(4)
R56 =
(44)
= I2 R34 .
0 0 1 0
0 0 0 i
(6)
The exchange matrix R34 cannot be obtained simply in this way because the quasiholes at 3
and 4 depend on the states of both qubits so that fusing either (1 , 2 ) or (5 , 6 ) would change
574
the state of the quasihole pair at (3 , 4 ). As we shall see in Section 7.2 this entanglement of
the two qubits allows us to construct immediately the Controlled-Z gate. Notice, however, that
(6)
R34 must be diagonal and therefore could be directly determined by simply using the OPE for
(6)
the fusion 3 4 alone. One way to see this is that if R34 were non-diagonal the exchange
of quasiholes at 3 and 4 would create a coherent superposition of the states I|NS and |NS
in the NeveuSchwartz sector (states with even number of fields acting on the vacuum belong
to the NS sector) which would be a violation of the superselection rule defined by the chiral
fermion parity F . In contrast, if there are odd number of s to the right of a and a+1 acting
on the vacuum then the fusion a a+1 generates I|R and |R, which are in the Ramond
sector where the chiral fermion parity is spontaneously broken [33], so that the above states could
(6)
(6)
indeed form coherent superpositions. This explains why R23 and R45 could be non-diagonal,
(6)
(6)
(6)
while R12
, R34 and R56 have to be diagonal. Thus, using the (neutral part of the) OPE for two
quasiholes, we have
1/8
|00 34
3 4
+ + + + ,
3/8
|10 34 + + + ,
3 4
3/8
|01 34 + + + ,
3 4
1/8
|11 34
3 4
+ + .
(45)
Therefore, the exchange 3 4 , which simply transforms 34 ei 34 , leads to

1/8
|00 ei 8 34
3 4
|01 iei 8 34 + + + ,
+ + + + ,
3/8
3 4
|10 iei 8 34 + + + ,
3/8
3 4
1/8
|11 ei 8 34
3 4
+ + .
1/8
Taking into account that there is another 34 , coming from the u(1) component of the quasihole
(6)
operator, we find the exchange matrix R34
in the basis (38) to be
1 0 0 0
0 i 0 0
(6)
=
R34
(46)
.
0 0 i 0
0 0 0 1
(6)
Unlike the other 6 quasiholes exchange matrices, R34 is not a factorized tensor product of the
4-quasiholes exchange matrices. Instead, there is an additional built-in structure in this representation of the braid group, which will eventually allow us to construct the Controlled-Z gate.
It is easy to check that the 6-quasiholes exchange matrices (41)(44) and (46) indeed satisfy
the Artin relations (33) for the braid group B6 [2], including the YangBaxter equations. As
mentioned before, using the Diminos algorithm [51] we can explicitly obtain the entire group
generated by the matrices (41), (42), (46), (43) and (44), giving the orders of the representation
of the braid group B6 and its monodromy subgroup

Image(B6 ) = 46 080,
Image(M6 ) = 32.
7.2. The Controlled-Z gate in terms of 6-quasiholes braidings
Using the explicit expressions for the 6-pt R-matrices, Eqs. (41), (46) and (44), it is straightforward to construct the most important two-qubit gatesthe CNOT or CZ gates in terms of the
575
Fig. 11. The braid diagram for the Controlled-Z gate realized by 3 commuting elementary 6-quasiparticle braidings
defined in Eq. (47). The symbol on the right is the standard quantum-computation notation for CZ.
braid matrices, e.g.,
1 0 0 0
0 1 0 0
(6) (6) 1 (6)
R34
R56 =
CZ = R12
.
0 0 1 0
0 0 0 1
Some insight into the CZ construction may be gained from the identity [24]
(47)
CZ = ei 4 ei 4 Z1 Z2 ei 4 Z1 ei 4 Z2 ,
where Z1 and Z2 are the Z gates acting on qubit 1 and 2 respectively. Because these matrices
are diagonal and square to I it is not difficult to prove that their exponents are actually equal
(6)
(6)
(up to overall phases) to the matrices R12 , R56 respectively, while the exponent of Z1 Z2 is
(6)
proportional to the inverse of R34 .
The braid diagram for the 6-quasihole exchanges corresponding to Eq. (47) is shown in
Fig. 11. In plotting Fig. 11 we have used that the three R-matrices entering Eq. (47) are diagonal and therefore commute, which also follows from the Artin relations (33), so that the
order of the exchanges is not important. Note the remarkable simplicity of this realization of the
CZ gatejust three elementary exchanges. This is one of the main advantages of the two-qubit
construction in terms of 6 quasiholes presented in Section 7.
7.3. Single-qubit gates in the two-qubit basis
Before we continue, it is important to show that we can efficiently express the single-qubit
gates into the two-qubit basis (38). The point is that the exchange matrices for 4 quasiholes,
which represent the single-qubit operations, belong to the braid group B4 while those for the
two-qubit gates are expressed in terms of braid matrices from the braid group B6 and the former
have different structure from the latter. Physically, the embedding of the one-qubit gates into
the two-qubit system is non-trivial because the entanglement creates non-local effects between
the two qubits. Nevertheless, the single-qubit construction in terms of 4-quasiholes exchanges is
certainly instructive for the representation of these gates in the two-qubit basis. For our purposes
it would be convenient to construct these gates explicitly.
The Hadamard gate acting on the first qubit can be expressed as
1 0 1
0
(6) 1 (6) 1 (6) 1 ei 4 0 1 0

1
H1 H I2 = R12
R23
R12
=
,
1 0 1 0
2
0 1 0 1
576
while that acting on the second qubit should be identified with
1 1 0 0
i 4
e 1 1 0 0
(6) (6) (6)
H2 I2 H = R56 R45 R56 =
.
2 0 0 1 1
0 0 1 1
(48)
Both Hadamard gates have similar structures to their single-qubit counterparts.

The phase gates acting on the first and second qubits are respectively
(6)
S1 = S I2 = R12
and
(6)
S2 = I2 S = R56 .
The embeddings of the other single-qubits into the two-qubit basis follow from these of H and
S because Z = S 2 and H ZH = X.
7.4. The Controlled-NOT gate
Now that we know how to construct the Controlled-Z gate, and how to embed the single-qubits
gates, entirely in terms of 6-quasiholes braidings, the CNOT gate is readily computed with the
help of the target-qubit Hadamard gate (48), i.e.,
1 0 0 0
0 1 0 0
1 1
CNOT = H2 CZH2 = R56 R45 R56
(49)
R34 R12 R45 R56
.
0 0 0 1
0 0 1 0
We used here that the target-qubit Hadamard gate could be executed by 3 exchanges, according
to Eq. (48), as well as the property (R56 )4 = I4 and some of the Artin relations (33) for the
6-quasiholes exchange matrices. An equivalent realization of CNOT, which gives precisely the
same result as in Eq. (49), could be given in terms of other 7 elementary exchanges corresponding
to the braid diagram in Fig. 12
1
1
R45 R34 R12 R56 R45 R34
.
CNOT = R34
(50)
This is the first known construction of the Controlled-Z and Controlled-NOT gates entirely in
terms of the braid matrices for 6 quasiholes in the Pfaffian TQC scheme, which certainly guarantees the exactness and topological protections of these gates.
Fig. 12. The braid diagram for the Controlled-NOT gate executed by 7 elementary 6-quasiparticle braidings corresponding to Eq. (50). The symbol on the right is the standard quantum-computation notation for CNOT.
577
= R2 R2 .
Fig. 13. The Controlled-Z gate precursor in terms of the monodromies CZ
15 16
Note that this construction of the CNOT gate is equivalent to the braid realization of the Bell
matrix of Refs. [26,54].
7.5. The BravyiKitaev Controlled-Z gate precursor
For the sake of completeness and comparison we shall also describe the existing idea [29]
to realize CZ by taking one quasihole, around two other, which suffers from the drawback that
the resulting transformation has one extra minus sign and thus has to be supplemented by external operations in order to produce the CZ gate, see below. Consider, e.g., the quasihole with
position 1 , from qubit 1 transported adiabatically around the two quasiholes, with positions 5
and 6 , of qubit 2 (or, equivalently, taking the two quasiholes comprising qubit 2 around one
quasihole of qubit 1) as shown in Fig. 13. This is obviously equivalent to first taking 1 around
6 and then around 5 so that this gate would be just the product of the two corresponding monodromies
1 0 0 0
0 1 0 0
= R 2 R 2 R 1 R 1 R 1 R45 R 2 R45 R34 R23 R12 =
CZ
(51)
.
15 16
56
12 23 34
0 0 1 0
0 0 0 1
The second equality in Eq. (51) is just an equivalent representation which can be readoff from
2
Fig. 13. We give for convenience also the explicit expression for R15
1 1 1
2
R15
= R12
R23 R34 R45 R34 R23 R12 = ei 4
0 0
0 0
0 1
1 0
0
1
0
0
1
0
,
0
0
directly. The braid diagram for

which together with Eq. (52) below can be used to compute CZ
this realization of the CZ gate precursor is shown in Fig. 14. Now it is obvious that the gate we
constructed in Eq. (51) differs from the CZ gate by having one more minus sign. The reason is
that if the second qubit is in the state |1 the transport of the quasihole at 1 will always produce
a minus sign whatever the state of the first qubit (note that the two states which are multiplied
by 1 by our gate (51) are exactly |01 and |11). The idea of BravyiKitaev [29] is to split the
first qubit into two 1/4-charge states only if this qubit is in the state |1 and then move the two
quasiholes (at positions 5 and 6 in our case) forming the second qubit around the first qubit
in
in order to execute this gate. This would remove the minus sign from the second row of CZ
Eq. (51) and, if successfully implemented, should give us a topologically protected Controlled-Z
gate.
578
Fig. 14. The braid diagram for the monodromy-based Controlled-Z gate precursor.
7.6. Realization of the BravyiKitaev two-qubit gate g3

One particular universal set of quantum gates, relevant for TQC with Pfaffian qubits, which
has been proposed by Bravyi and Kitaev [29]. To construct the gate g3 it would be instructive
2 as taking around . As is obvious from Fig. 15, this
to compute first the monodromy R16
1
6
monodromy can be expressed in terms of the elementary 6-quasiholes exchanges5 (omitting the
superscript (6) in the R-matrices) as
0 0 0 1
0 0 1 0
1 1 1 1 2
2
R16
(52)
= R12
R23 R34 R45 R56 R45 R34 R23 R12 =
.
0 1 0 0
1 0 0 0
Then the exchange R16 is just the square-root of Eq. (52), i.e.,
1 1 1 1
R16 = R12
R23 R34 R45 R56 R45 R34 R23 R12 .
(53)
The direct computation, using the explicit formulas, Eqs. (41), (42), (46), (43) and (44) for the
elementary R matrices, shows that the two-qubit gate g3 identical with R16 , i.e.,
1
0
0 i
1 0
1 i 0
R16
g3 .
0
i
1
0
2
i 0
0
1
The braid diagram for the realization of R16 is shown in Fig. 16.
Remark 4. It is worth stressing that the TQC scheme of Das Sarma et al. [28] is essentially
based on the monodromy transformations of the multi-quasihole Pfaffian wave functions. The
only exception is the BravyiKitaev gate g3 , which was proposed to be constructed schematically
in Ref. [29] by braiding two among 4 quasiholes (not by braiding 6 quasiholes as it should be).
Note, however, that as a two-qubit gate acting on Pfaffian qubits, g3 must be constructed in terms
of transformations of the 6-quasiholes states, which has not been done in Ref. [29]. In particular,
it is not clear from the 4-quasihole braiding construction of Ref. [29], which two quasiholes
among the 6 ones must be exchanged in order to obtain the gate g3 , and whether this is at all
possible. The first explicit results proving the usefulness of braidings for topological quantum
computation with Pfaffian quasiholes have been obtained in Ref. [47] and Eq. (53) is the braiding
implementation of g3 .
5 Note that the clockwise exchanges correspond to the inverse exchange matrices.
579
2 executed by taking around in a counter-clockwise direction.

Fig. 15. The monodromy R16
1
6
Fig. 16. Implementing the two-qubit gate g3 by counter-clockwise exchange of the quasiholes with positions 1 and 6 .
7.7. The non-demolition measurement gate

One of the quantum gates in the universal TQC schemes reviewed in Ref. [29] is executed by
a non-demolition measurement of the total topological charge of two qubits. According to our
construction of the two-qubit states this measurement is equivalent to the transformation
|00 |00,
|01 |01,
|10 |10,
|11 |01.
Therefore we can identify the non-demolition measurement gate with the monodromy matrix
corresponding to the adiabatic transport of 3 around 4
1 0
0 0
(6) 2 0 1 0 0
R34 =
Z1 Z2 ,
0 0 1 0
0 0
0 1
(6) 2
(6) 2
where Z1 = (R12
) and Z2 = (R56
) are the Pauli Z-gates over the first and second qubits, respectively. While the non-demolition measurement of the total topological charge might happen
to be noisy, the implementation of the above quantum gate as a 6-quasiholes state monodromy is
completely protected by topology.
7.8. The two-qubit SWAP gate

Once we now how to construct the CNOT gate it is straightforward to obtain the two-qubit
Swap gate in terms of three CNOTs [24] as shown in the top line of Fig. 17. Here we shall
demonstrate that it is possible to implement the SWAP gate with 15 elementary exchanges.
The bottom line of Fig. 17 shows how to express the swapped CNOT in terms of the CZ and
(4)
Hadamard gates. It turns out that in this circuit H is essentially equivalent to R23 so that sub(6)
(6)
stituting (H I2 ) R23 and (I2 H ) R45 , as well as using Eq. (47) for CZ, we finally
580
Fig. 17. The two-qubit SWAP gate realized by three CNOTs.
obtain
SWAP (I2 H )CZ(I2 H )(H I2 )CZ(H I2 )(I2 H )CZ(I2 H )
1
1
1
R45 R34
R12 R56 R45 R23 R34
R12 R56 R23 R45 R34
R12 R56 R45
1 0 0 0
0 0 1 0
=i
.
0 1 0 0
0 0 0 1
(54)
8. Universal TQC scheme based on the Hadamard gate H , phase gate S, CNOT and
Toffoli gate
One of the standard universal set of gates, which could be used for universal quantum computation, includes the Hadamard gate H , the phase gate S, the two-qubit CNOT gate and the
Toffoli gate, which is a three-qubit ControlledControlled-NOT (CCNOT) gate [24]. In order to
execute three-qubit gates, such as the Toffoli and Fredkin gates, we need to consider a system
with 8 quasiholes, whose Hilbert subspace of states (for fixed positions of the quasiholes) has
8
dimension 2 2 1 = 8. Here we shall assume that the third qubit is defined by 2 more quasiholes
at positions 7 and 8 as shown in Fig. 18. Then the three-qubit basis can be written in terms of
the Ising spin fields as follows
|000 + + + + + + + + ,
|001 + + + + + + ,
|010 + + + + + + ,
|011 + + + + + + ,
|100 + + + + + + ,
|101 + + + + + + ,
|110 + + + + + + ,
|111 + + + + .
(55)
Now the fermion parity conservation requires that

e3 e 4 = e 1 e 2 e 5 e 6 e 7 e 8 ,
which reduces the number of independent states to from 16 to 8. Although we have chosen the
quasiholes at 3 and 4 in such a way to preserve the fermion parity, any other choice of the
positions of quasiholes representing the three qubits would be equivalent to Eq. (55) because the
braid matrices for the elementary exchanges would be just conjugated by an element of the braid
group, and the Artin relations (33) are invariant under conjugation.
581
Fig. 18. Three qubits constructed from 8 quasiholes.
8.1. Exchange matrices for 8 quasiholes

Using the fusion rules of the non-Abelian quasiholes we can express the exchange matrices
for 8 quasiholes recursively in terms of those for 6 quasiholes as follows:
(8)
(6)
R12 = diag(1, 1, 1, 1, i, i, i, i) = R12 I2 ,
1
0
0
0 i 0
1
0
0
0 i
0
0
0
1
0
0
0
ei 4
0
0
1
0
0
0
(8)
R23 =
i 0
0
0
1
0
2
0
0
1
0 i 0
0
0 i 0
0
0
0
0
0 i 0
0
0
0
i
0
0
0
1
0
0
0
i
(6)
= R23 I2 ,
0
0
1
(8)
R34 = diag(1, i, i, 1, i, 1, 1, i),
1
0 i 0
1
0 i
0
i
0
1
0
ei 4
1
0 i 0
(8)
R45 =
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
i
0
0
0
0
i
0
1
0
0
0
0
(6)
= R45 I2 ,
0
0
1
R56 = diag(1, 1, i, i, 1, 1, i, i) = R56 I2 ,
1
0
0 i 0
0
1 i 0
0
0
0
0
i
1
0
0
0
ei 4
0
1
0
0
i 0
(8)
R67 =
0
0
0
0
1
0
2
0
0
0
0
1
0
0
0
0
0
0 i
0
0
0
0 i 0
0
0
0
0
0
i
1
0
0
0
0
,
i
0
1
(8)
(8)
0
0
0
0
1
0
i
0
(6)
(6)
R78 = diag(1, i, 1, i, 1, i, 1, i) = I2 R56 .
(56)
(57)
(58)
(59)
(60)
(61)
(62)
It is not difficult to check that the exchange matrices (56)(61) and (62) satisfy the Artin relations (33) for the braid group B8 . Again, the order of the representation of the braid group B8
and its monodromy subgroup can be obtained by Diminos algorithm [51] to be

Image(B8 ) = 5 160 960,
Image(M8 ) = 128.
582
As an illustration of the derivation of the 8-quasiholes exchange matrices, let us compute the last
(8)
row of R67 , i.e., we consider the transformation of the state |111 when we exchange 6 with 7 .
Because the state of the second and the third qubits is independent of the quasiholes at 1 and
2 , we could fuse 1 2 obtaining in this way a 6-quasiholes state whose braiding properties
are already known. Indeed, using the OPE for two quasiholes we find

12
(2 )+ (3 ) (4 )+ (5 ) (6 )+ (7 ) (8 )
|111
1 2
2

12
1

(3 ) (4 )+ (5 ) (6 )+ (7 ) (8 ) |01,
2 3
2 223
where we used the OPE [46] (2 )e3 (3 ) (223 )1/2 e3 (3 ), for 2 3 , and the identity
+ + + + + + |01. It is now obvious that the exchange of 6 with
7 in the three-qubit state |111 is equivalent to the exchange of the fifth and sixth quasiholes in
(6)
the state |01 so that, taking R45 from Eq. (43), we obtain

(6)

R45 ei 4
12
1
|111
i|00 + |01
2 223
2

1
ei 4 12
i + (3 )+ (4 )+ (5 )+ (6 )+ (7 )+ (8 )
=
2 223
2

+ + (3 )+ (4 )+ (5 ) (6 )+ (7 ) (8 )

ei 4 12
i (2 ) (3 )+ (4 )+ (5 )+ (6 )+ (7 )+ (8 )

2 3
2
2

+ (2 ) (3 )+ (4 )+ (5 ) (6 )+ (7 ) (8 )
1 2

ei 4
i + (1 ) (2 )+ (3 ) (4 )+ (5 )+ (6 )+ (7 )+ (8 )
2

+ + (1 ) (2 )+ (3 ) (4 )+ (5 ) (6 )+ (7 ) (8 )

ei 4
i|100 + |111 ,
2
(63)
(8)
which exactly reproduces the last row of R67 . In the above derivation we restored the 8quasiholes states using the same OPEs for 2 3 and 1 2 , however in reverse, as well as
used the identity (3 )+ (4 ) + (3 ) (4 ).
Remark 5. Due to the specifics of the braid group representation, it may not be always possible
to represent exactly the single- and two-qubit gates in the three-qubit basis (55). Indeed, the
(6)
6-quasiholes exchange matrix R34
, defined in Eq. (46), is not a factorized tensor product of
the exchange matrices for 4 quasiholes, rather it contains the built-in CZ matrix (47). Therefore
some tensor products, which are trivial otherwise, might not be constructed directly in terms of
the elementary exchange matrices for 8 quasiholes. One consequence of this peculiarity is that
some one-qubit and two-qubit gates would be easier realizable in the three-qubit basis (55) in
terms of elementary 8-quasiholes exchange matrices, however, up to a pair of extra minus signs,
see Section 8.2. While the exact construction of the one-qubit and two-qubit gates would require
more work, their simplified versions might be sufficient in most cases.
583
8.2. Embedding of one-qubit and two-qubit gates into a three-qubit system

The three one-qubit phase gates are directly expressed as single elementary 8-quasiholes exchange matrices, i.e.,
(8)
S1 S I4 = R12 ,
(8)
S2 I2 S I2 = R56 ,
(8)
S3 I4 S = R78 .
The first one-qubit Hadamard gate can be constructed exactly in terms of the exchange matrices
for 8 quasiholes by (skipping the superscript (8) of R)
1 0 0 0 1
0
0
0
1
0
0
0 1 0 0 0
0
0
1
0
0
0
1
0
i 4
e
0
0
1
0 0 0 1 0
1 1 1
H1 = H I4 R12
(64)
R23 R12 =
,
1 0 0 0 1 0
0
0
2
0
0 1 0 0 0 1 0
0 0 1 0 0
0 1 0
0 0 0 1 0
0
0 1
while the second Hadamard gate could be constructed as follows
1 0 1
0
1
0 1 0
1
0
1
0
ei 4
0 1 0 1
1 1 1
H2 I2 H I2 R56 R45 R56 =
0 0 0
0
2
0
0 0 0
0 0 0
0
0 0 0
0
0
0
0
0
1
0
1
0
0 0
0 0
0 0
0 0
0 1
1 0
0 1
1 0
0
0
0
. (65)
0
0
1
The third Hadamard gate could also be reproduced up to some swapping by

1 1 1 1 1 1 1
R45 R56 R67 R56 R45 R78
H3 = I4 H R78
1 1
0 0 0 0
0
0
1 1 0 0 0 0
0
0
1
1
0
0
0
ei 4
1 1 0 0
0
0 0
=
0
0
0
0
1
1
0
2
0 0 1 1 0
0 0
0 0
0 0 0 0 1
0 0
0 0 0 0
1
0
0
0
.
0
1
1
(66)
2 ,
It might be useful to give also the realization of the single-qubit NOT gates: X1 X I4 = R23
2
2
2
X2 I2 X I2 = R45 and X3 I4 X = R45 R67 .
We should stress again that the topological entanglement between the qubits mentioned in
Section 6 leads to serious difficulties for efficient embedding of the one-qubit and two-qubit
gates in three-qubit systems. For example, the NOT gate X3 acting on the third qubit has a
different structure than just a tensor product of the exchange matrix producing the single-qubit X
gate. Similarly, the Hadamard gates acting on the first and second qubits have slightly different
structures from their single-qubit counterpart H , while that acting on the third qubit cannot even
be obtained exactly with the same number of elementary exchanges as H . This seems to be a
584
common problem arising in all TQC schemes using non-Abelian anyons in FQH systems, whose
general solution is still missing. Moreover, it appears that the two-qubit Controlled-NOT gates in
a three-qubit systems cannot be directly constructed in the three-qubit basis (55) because of the
topological entanglement between the two qubits and the third one. This requires more work and
will be reported elsewhere. Just for reference, we give a simple implementation of a three-qubit
operation which is very close to CNOT2 = I2 CNOT
2 I2 CNOT R 1 R56 SWAP
2 R36 R45 SWAP
2
CNOT
12
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0
=
,
0 0 0 0 0 1 0 0
0 0 0 0 1 0 0 0
0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 1
(67)
1 1 1
2 is defined below.
R45 R34 R45 R56 and the gate SWAP
where R36 = R56
The two-qubit SWAP gates can be simply realized by braiding in the three-qubit basis (55)
(up to overall phases and pairs of extra minus signs) by
1 SWAP I2 R45 R 1 R 1 R45 R23 R 1 R 1 R23 R45 R 1 R 1 R45

SWAP
56 34
34 12
56 34
1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0
0 0 0 1 0 0 0
0
0 0 0 0 1 0 0
0
=
,
0 1 0 0 0 0 0
0
0 0 1 0 0 0 0
0
0
0 0 0 0 0 1 0
0
0 0 0 0 0 0 1
1 0 0 0 0 0 0
0 0 1 0 0 0 0
0 1 0 0 0 0 0
0 0 0 1 0 0 0
1
1
2 I2 SWAP R67 R R R67 =
SWAP
78 56
0 0 0 0 1 0 0
0 0 0 0 0 0 1
0 0 0 0 0 1 0
0 0 0 0 0 0 0
0
0
0
.
0
0
1
(68)
The SWAP gates turn out to be very important because they can be used to construct gates acting
on one of the qubits in terms of similar gates acting on another qubit. For example, the CNOT
gate acting on the first and the third qubits can be expressed in terms of the CNOT acting on the
first and the second qubits plus two gates SWAP2 (see Fig. 19). The extra minus signs appearing
in some of the three-qubit operations are not an innocent thing, because they may have different
properties from the standard gates. However, it appears that in many cases these simplified gates,
which are much simpler to construct, can be used instead of the standard once.
The three-qubit Toffoli gate [24] can be constructed in terms of the Controlled-S gate
and CNOT like in Ref. [47] using the relation between the Toffoli gate and the Controlled
585
Fig. 19. The CNOT acting on qubits 1 and 3 expressed in terms of CNOT I2 , which is the CNOT gate acting on qubits
1 and 2, and two gates I2 SWAP.
Controlled-Z gate or by a braid-group based ControlledControlled-Z gate precursor Ref. [47],

which must be supplemented by the BravyiKitaev construction [47].
9. Discussion
In this paper we explicitly implemented all single-qubit gates in the Pfaffian TQC scheme,
except for the /8 one, in terms of 4-quasihole braidings, as well as the two-qubit Controlled-Z
and CNOT gates in terms of 6-quasihole braidings. These gates, which are known to form a Clifford group, are realized in a completely topologically protected way because of the topological
nature of the braid operations in the FQH liquids. This work is an extension of the topological
quantum computation scheme of Ref. [28] using pairs of Pfaffian quasiholes localized on antidots to construct elementary qubits and execute logical NOT on them. While the original TQC
scheme of Ref. [28] used only monodromy transformations to realize quantum gates, we, for
the first time, exploited explicitly quasihole braiding in the Pfaffian FQH state to construct the
single-qubit Hadamard gate H , the phase gate S and the CNOT gate. Although the Gottesmann
Knill theorem says that any circuit based only on the Clifford group gates could be efficiently
simulated on a (probabilistic) classical computer these gates do play a crucial role in quantum
computation, especially in the error-correcting algorithms [24].
Due to the topological entanglement between the separate qubits realized by non-Abelian
anyons in FQH systems some difficulties arise when trying to embed the one-qubit and two-qubit
gates into systems with more qubits. This makes the embedding of Clifford gates non-trivial and
requires more work.
For implementing three-qubit gates such as the Toffoli and Fredkin gates [24], in the Pfaffian TQC scheme, we considered Pfaffian wave functions with 8-quasiholes, in which case the
topological degeneracy of the space of correlation functions is 8 [12,38]. We derived explicitly
the braid matrices for the elementary 8-quasiholes exchanges, which serve as building blocks for
constructing all three-qubit gates. More work in this direction, including eventually the topologically protected construction of the Toffoli gate would be reported elsewhere.
To conclude, let us make some remarks about the possible observation of the non-Abelian
statistics. We believe that in order to observe the Pfaffian phase at = 5/2 one should perform
the experiment at temperature below 15 mK. The point is that there might exists another incompressible Abelian phase, which was called the Extended Pfaffian (EPf) state [33], that could
also be realized at = 5/2. Perhaps the most observable difference between the two phases is
in the electric charge of their quasiparticles: 1/4 for the Pfaffian and 1/2 for the EPf. The EPf
phase was obtained mathematically by a local chiral algebra extension of the Pfaffian state and
satisfies all conditions necessary for an incompressible quantum Hall state [33]. The motivation
for introducing this new state is that there is a persistent unexplained kink around Tc = 15 mK
observed in the thermal activation experiment [55] showing two different slops that presumably
586
correspond to two different gaps below and above the critical temperature. Analyzing the edge
states CFT it has been proposed in Ref. [33] a possible explanation of the kink in terms of a
finite temperature two-step phase transition between the Pfaffian and the EPf state involving an
intermediate compressible state of composite fermions. Here is a brief description of the process
(see Section 9 in Ref. [33] for more details): at low temperature the system is definitely in the
Pfaffian phase, as the numerical calculations suggest. As temperature increases to about 1/2 of
the Pfaffian-phase gap, which was estimated to be about 33 mK (for the sample of Ref. [55]),
the system becomes more and more compressible (look at the behavior of the free energy on the
edge, Fig. 5 in Ref. [33]) leading to a second order phase transition to the compressible state of
composite fermions (which has the same topological structure like the Pfaffian state but having
at the same time the Z2 symmetry of the EPf state that is broken spontaneously in the Pfaffian
phase). Immediately after that, as temperature continues to increase, there is a first order phase
transition to the EPf state which is expected to have a higher gap than the Pfaffian state.
Acknowledgements
I thank Ivan Todorov, Ady Stern, Valentina Petkova, Chetan Nayak, Lyudmil Hadjiivanov
and Michael Geller for many helpful discussions. This work has been partially supported by
the FP5-EUCLID Network Program of the European Commission under Contract No. HPRNCT-2002-00325 and by the Bulgarian National Council for Scientific Research under Contract
No. F-1406.
Appendix A. Binomial series expansion of the 4-pt function and analytic continuation
In this appendix we shall give some details about the analytic continuation of the function (19),
which might be necessary for the understanding of the results of Section 3. Using the standard
complex analysis notion [56] of a branching point as a multi-valued isolated singular point, we
consider a punctured neighborhood of the branching point, denoted as U in which we would
like to continue the element (U, f ) of the function (19) from the simply-connected sub-domain
U U along any path. For example, for the branching point at z = 0 we can choose

U = z 0 < |z| < 1 ,
U = z |z 1/2| < 1/2 ,
as shown in Fig. A.1. Then the analytic continuation along the contour

1

0 = z z = eit , 0 t 2
2
(A.1)
changes the sign in front of the inner root but not the one of the outer root because 0 does not
encircle z = 1. This can be verified directly by using the (fractional powers) Laurent expansion,
which in this case can be obtained by the binomial series expansion
(1 + x) =
n=0
( + 1)
xn,
(n + 1)( + 1 n)
|x| < 1,
applied for = 1/2. Using the defining function property (z) = (z 1)(z 1), we get

1
1
1
(A.2)
1 z= 1
z z z z , z U .
2
8
16
587
Fig. A.1. (Color online.) Domains and contours for the different values of |z| used for the analytic continuation and
binomial series expansion.

That is why 1 z 1 z, for |z| < 1, when continuing z e2i z along any contour

in U , which is homotopic to (A.1).
For the branching point at z = 1, on the other hand, we consider the domains V and V shown
again in Fig. A.1 defined by

V = z |z 1/2| < 1/2 .
V = z 0 < |z 1| < 1 ,
Then the analytic continuation of the element (V , f ) of the function (19) from V to V along any
contour in V homotopic to

1 it

1 = z z = 1 + e , 0 t 2
2
does not change the inner root sign because the point z = 0 is outside the contour. It only
changes the sign of the outer root whenever the inner root sign is . Indeed, representing
(z1)2
z = 1 + (z 1) (1 + z1
+ (z1)
2
8
16 ) for |z 1| 1, and using again the binomial
expansion in V we get

z1
5
2
1+ z= 2 1+
(z 1) + ,
(A.3)
8
128
z1
7
z1
(A.4)
1 z=
+
(z 1)2 + , z V .
1
8
128
i 2
of Eq. (A.4) but not in Eq. (A.3) implies that

The appearance of z 1 in the right-hand
side
when
z
encircles
the
point
1
the
function
1
z acquires one minus sign, while the function

1 + z is single-valued.
Recall that for a contour passing through the points z = 0 or z = 1 it is not possible to make
analytic continuation.
588
Finally for |z| > 1 we use W = {z | 1 < |z| < }, W = {z | |z 2| < 1}, and the continuation
along the contour

2 = z z = 2eit , 0 t 2
changes both the sing of the inner root and that of the outer root when the inner sign is .
Because in this case |1/ z| < 1 the binomial expansion with respect to 1/ z gives

1
1
1
1 z= 4 z 1
, 1 < |z| < ,

2 z 8z 16z z
and that explicitly shows that z = is a branching point of order 4.
That is how we conclude that the general contour , which is shown on Fig. A.1, corresponding to the read-out transformation (17) is homotopic to 0 or 0 1 depending on the value
of |x|, i.e.,

0
for |x| < 1,

0 1 for |x| > 1,
which explains once again the analytic continuation results obtained in Section 3.
References
[1] F. Wilczek, Fractional Statistics and Anyon Superconductivity, World Scientific, Singapore, 1990.
[2] J.S. Birman, Braids, Links and Mapping Class Groups, eighty second ed., Annals of Mathematics Studies, Princeton
Univ. Press, 1974.
[3] K. Fredenhagen, K.H. Rehren, B. Schroer, Superselection sectors with braid group statistics and exchange algebras,
Commun. Math. Phys. 125 (1989) 201.
[4] J. Frhlich, F. Gabbiani, Braid statistics in local quantum theory, Rev. Math. Phys. 2 (1990) 251.
[5] K. Fredenhagen, M.R. Gaberdiel, S.M. Rueger, Scattering states of plektons (particles with braid group statistics)
in 2 + 1 dimensional quantum field theory, Commun. Math. Phys. 75 (1996) 319, hep-th/9410115.
[6] J. Eisenstein, K. Cooper, L. Pfeiffer, K. West, Insulating and fractional quantum Hall states in the first excited
Landau level, Phys. Rev. Lett. 88 (2002) 076801.
[7] J. Xia, W. Pan, C. Vicente, E. Adams, N. Sullivan, H. Stormer, D. Tsui, L. Pfeiffer, K. Baldwin, K. West, Electron
correlation in the second Landau level: A competition between many nearly degenerated quantum phases, Phys.
Rev. Lett. 93 (2004) 176809.
[8] N. Read, D. Green, Paired states of fermions in two dimensions with breaking of parity and time-reversal symmetries
and the fractional quantum Hall effect, Phys. Rev. B 61 (2000) 10267, cond-mat/9906453.
[9] D. Ivanov, Non-Abelian statistics of half-quantum vortices in p-wave superconductors, Phys. Rev. Lett. 86 (2001)
268.
[10] R. Morf, Transition from quantum Hall to compressible states in the second Landau level: New light on the = 5/2
enigma, Phys. Rev. Lett. 80 (1998) 1505, cond-mat/9809024.
[11] E. Rezayi, F. Haldane, Incompressible paired Hall state, strip order and composite fermion liquid phase in half-filled
Landau level, Phys. Rev. Lett. 84 (2000) 4685, cond-mat/9906137.
[12] G. Moore, N. Read, Nonabelions in the fractional quantum Hall effect, Nucl. Phys. B 360 (1991) 362.
[13] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1997.
[14] A. Cappelli, L.S. Georgiev, I.T. Todorov, A unified conformal field theory approach to paired quantum Hall states,
Commun. Math. Phys. 205 (1999) 657689, preprint ESI-621 (1998);
A. Cappelli, L.S. Georgiev, I.T. Todorov, hep-th/9810105.
[15] R. de Picciotto, M. Reznikov, M. Meiblum, V. Umansky, G. Bunin, D. Mahalu, Nature (London) 389 (1997) 162.
[16] A. Stern, B.I. Halperin, Proposed experiments to probe the non-Abelian = 5/2 quantum Hall state, Phys. Rev.
Lett. 94 (2006) 016802, cond-mat/0508447.
[17] P. Bonderson, A. Kitaev, K. Shtengel, Detecting non-Abelian statistics in the = 5/2 fractional quantum Hall state,
Phys. Rev. Lett. 96 (2006) 016803, cond-mat/0508616.
[18] L.S. Georgiev, M.R. Geller, AharonovBohm effect in the non-Abelian quantum Hall fluid, Phys. Rev. B 73 (2006)
205310, cond-mat/0511236.
589
[19] D.E. Feldman, A. Kitaev, Detecting non-Abelian statistics with electronic MachZehnder interferometer, Phys. Rev.
Lett. 97 (2006) 186803, cond-mat/0607541.
[20] P. Bonderson, K. Shtengel, J.K. Slingerland, Probing non-Abelian statistics with two-particle interferometry, Phys.
Rev. Lett. 97 (2006) 016401, cond-mat/0601242.
[21] S.B. Chung, M. Stone, Proposal for reading out anyon qubits in non-Abelian = 12/5 quantum Hall state, Phys.
Rev. B 73 (2006) 245311, cond-mat/0601594.
[22] N. Read, E. Rezayi, Beyond paired quantum Hall states: Parafermions and incompressible states in the first excited
Landau level, Phys. Rev. B 59 (1998) 8084.
[23] A. Cappelli, L.S. Georgiev, I.T. Todorov, Parafermion Hall states from coset projections of Abelian conformal
theories, Nucl. Phys. B 599 (2001) 499530, hep-th/0009229.
[24] M. Nielsen, I. Chuang, Quantum Computation and Quantum Information, Cambridge Univ. Press, 2000.
[25] A. Kitaev, Fault-tolerant quantum computation by anyons, Ann. Phys. (N.Y.) 303 (2003) 2.
[26] L.H. Kauffman, S.J. Lomonaco Jr., Braiding operators are universal quantum gates, New J. Phys. 6 (2004) 134,
quant-ph/0401090.
[27] J. Preskill, Topological Quantum Computation, in: Lecture Notes for Physics, vol. 219, 2004, http://www.theory.
caltech.edu/~preskill/ph219.
[28] S.D. Sarma, M. Freedman, C. Nayak, Topologically-protected qubits from a possible non-Abelian fractional quantum Hall state, Phys. Rev. Lett. 94 (2005) 166802, cond-mat/0412343.
[29] M. Freedman, C. Nayak, K. Walker, Towards universal topological quantum computation in the = 5/2 fractional
quantum Hall state, Phys. Rev. B 73 (2006) 245307, cond-mat/0512066.
[30] J. Frhlich, U.M. Studer, E. Thiran, A classification of quantum Hall fluids, J. Stat. Phys. 86 (1997) 821, condmat/9503113.
[31] J. Frhlich, C. King, Two-dimensional conformal field theory and three-dimensional topology, Int. J. Mod. Phys.
A 4 (1996) 5321.
[32] J. Frhlich, B. Pedrini, C. Schweigert, J. Walcher, Universality in quantum Hall systems: Coset construction of
incompressible states, J. Stat. Phys. 103 (2001) 527, cond-mat/0002330.
[33] L.S. Georgiev, The = 5/2 quantum Hall state revisited: Spontaneous breaking of the chiral fermion parity and
phase transition between Abelian and non-Abelian statistics, Nucl. Phys. B 651 (2003) 331360, hep-th/0108173.
[34] Y. Zhang, L.H. Kauffman, R.F. Werner, Permutation and its partial transpose, quant-ph/0606005.
[35] H. Bombin, M. Martin-Delgado, Topological quantum distillation, Phys. Rev. Lett. 97 (2006) 180501, quantph/0605138.
[36] D.V. Averin, V.J. Goldman, Quantum computation with quasiparticles of the fractional quantum Hall effect, Solid
State Commun. 121 (2002) 25, cond-mat/0110193.
[37] J. Preskill, Quantum Information and Computation, Lecture Notes for Physics, vol. 229, 1998, http://www.theory.
caltech.edu/~preskill/ph229.
[38] C. Nayak, F. Wilczek, 2n quasihole states realize 2n1 -dimensional spinor braiding statistics in paired quantum
Hall states, Nucl.Phys. B 479 (1996) 529, cond-mat/9605145.
[39] J.K. Slingerland, F.A. Bais, Quantum groups and non-Abelian braiding in quantum Hall systems, Nucl. Phys. B 612
(2001) 229290, cond-mat/0104035.
[40] S. Bravyi, A. Kitaev, Universal quantum computation with ideal Clifford gates and noisy ancillas, Phys. Rev. A 71
(2005) 022316.
[41] A. Stern, F. von Oppen, E. Mariani, Geometric phases and quantum entanglement as building blocks for non-Abelian
quasiparticle statistics, Phys. Rev. B 70 (2004) 205338.
[42] V. Goldman, Quantum antidot as an electrometer: Observation of fractional charge, Physica E 1 (1997) 1520.
[43] C. Nayak, private communication, 2005.
[44] Y. Ji, Y. Chung, D. Sprinzak, M. Heiblum, D. Mahalu, H. Shtrikman, An electronic MachZehnder interferometer,
Nature 422 (2003) 415.
[45] B.J. Overbosch, F.A. Bais, Inequivalent classes of interference experiments with non-Abelian anyons, Phys. Rev.
A 64 (2001) 062107.
[46] P. Furlan, G. Sotkov, I. Todorov, Two-dimensional conformal quantum field theory, Riv. Nuovo Cimento 12 (6)
(1989) 1.
[47] L.S. Georgiev, Topologically protected gates for quantum computation with non-Abelian anyons in the Pfaffian
quantum Hall state, Phys. Rev. B 74 (2006) 235112, cond-mat/0607125.
[48] D. Deustch, A. Ekert, R. Lupacchini, Machines, logic and quantum physics, math.HO/9911150.
[49] H.S.M. Coxeter, W.O.J. Moser, Generators and Relations for Discrete Groups, third ed., Springer, Berlin, 1972.
[50] M. Hamermesh, Group Theory and Its Application to Physical Problems, Dover Books on Physics and Chemistry,
1989.
590
[51]
[52]
[53]
[54]
G. Butler, Diminos algorithm, Lecture Notes Comput. Sci. 559 (1991) 13.
N. Read, Non-Abelian braid statistics versus projective permutation statistics, J. Math. Phys. 44 (2003) 558.
D. Aharonov, A simple proof that Toffoli and Hadamard are quantum universal, quant-ph/0301040.
Y. Zhang, L.H. Kauffman, M.-L. Ge, Yang-baxterizations, universal quantum gates and Hamiltonians, Quant. Inf.
Proc. 4 (2005) 159, quant-ph/0502015.
[55] W. Pan, J.-S. Xia, V. Shvarts, D.E. Adams, H.L. Strmer, D.C. Tsui, L.N. Pfeiffer, K.W. Baldwin, K.W. West, Exact
quantization of the even-denominator fractional quantum Hall state at = 5/2 Landau level filling factor, Phys.
Rev. Lett. 83 (1999) 3530, cond-mat/9907356.
[56] B. Shabat, Introduction to Complex Analysis, Nauka, Moscow, 1969 (in Russian).
Multiple reference states and complete spectrum

of the Zn Belavin model with open boundaries
Wen-Li Yang a,b, , Yao-Zhong Zhang b
a Institute of Modern Physics, Northwest University, Xian 710069, PR China
b Department of Mathematics, University of Queensland, Brisbane, QLD 4072, Australia

Abstract
The multiple reference state structure of the Zn Belavin model with non-diagonal boundary terms is
discovered. It is found that there exist n reference states, each of them yields a set of eigenvalues and
Bethe Ansatz equations of the transfer matrix. These n sets of eigenvalues together constitute the complete
spectrum of the model. In the quasi-classic limit, they give the complete spectrum of the corresponding
Gaudin model.
PACS: 75.10.Pq; 04.20.Jb; 05.50.+q
Keywords: Algebraic Bethe Ansatz; Zn Belavin model; Gaudin model
1. Introduction
Our understanding of quantum phase transitions and critical phenomena has been greatly
enhanced by the study of exactly solvable models (integrable models) [1]. Such exact results
provide valuable insights into the key theoretical development of universality classes in areas
ranging from modern condensed physics [2] to string and super-symmetric YangMills theories [3]. Among solvable models, elliptic ones stand out as a particularly important class due to
the fact that most other ones can be obtained from them by some trigonometric or rational limits.
* Corresponding author at: Department of Mathematics, University of Queensland, Brisbane, QLD 4072, Australia.
E-mail address: wenli@maths.uq.edu.au (W.-L. Yang).

doi:10.1016/j.nuclphysb.2007.07.024
592
W.-L. Yang, Y.-Z. Zhang / Nuclear Physics B 789 [FS] (2008) 591609
In this paper, we focus on the elliptic Zn Belavin model [4] with integrable boundary conditions,
with the celebrated XYZ spin chain as the special case of n = 2.
Two-dimensional integrable models have traditionally been solved by imposing periodic
boundary conditions. For such bulk systems, the quantum YangBaxter equation (QYBE)
R12 (u1 u2 )R13 (u1 u3 )R23 (u2 u3 ) = R23 (u2 u3 )R13 (u1 u3 )R12 (u1 u2 )
(1.1)
leads to families of commuting row-to-row transfer matrix which may be diagonalized by the
quantum inverse scattering method (QISM) (or algebraic Bethe Ansatz) [5].
Not all boundary conditions are compatible with integrability in the bulk. The bulk integrability is only preserved when one imposes certain boundary conditions. In [6], Sklyanin developed
the boundary QISM, which may be used to describe integrable systems on a finite interval with
independent boundary conditions at each end. This boundary QISM uses a new algebraic structure, the reflection equation (RE) algebra. The solutions to the RE and its dual are called boundary
K-matrices which in turn give rise to boundary conditions compatible with the integrability of
the bulk model [68].
The boundary QISM has been successfully applied to diagonalize the double-row transfer
matrices of various integrable models with non-trivial boundary conditions mostly corresponding
to the diagonal K-matrices. The problem of diagonalizing the double-row transfer matrix for
most general non-diagonal K-matrices by the algebraic Bethe Ansatz has been long-standing due
to the difficulty of finding suitable reference states (or pseudo-vacuum states). Recently, much
progress has been made for the open XXZ spin chain. Bethe Ansatz solutions for non-diagonal
boundary terms where the boundary parameters obey some constraints have been proposed by
various approaches1 [916]. It has been found that in order to obtain the complete spectrum of the
model two sets of Bethe Ansatz equations and consequently two sets of eigenvalues are needed
[15,19], in contrast with the diagonal boundary case [6]. This suggests that in the framework of
algebraic Bethe Ansatz there should exist two reference states corresponding to the two sets of
Bethe Ansatz equations and eigenvalues. Such multiple reference state structure was confirmed
in our recent work [20].
However, for models related to higher rank algebras [2126] only one reference state, and
consequently only one set of eigenvalues and Bethe Ansatz equations of their transfer matrices,
have been constructed so far. It is natural to expect that there exist extra reference states, giving
rise to extra sets of eigenvalues and associated Bethe Ansatz equations for such models. In this
paper we investigate the multiple reference state structure for an elliptic model related to An1
algebrathe Zn Belavin model with general non-diagonal boundary terms. It is found that there
actually exist n reference states for such a model. Each of these reference states yields a set
of eigenvalues and corresponding Bethe Ansatz equations of the transfer matrix. The n sets of
eigenvalues together constitute the complete spectrum of the model. In the quasi-classic limit,
they give the corresponding spectrum of the associated Gaudin model [27].
The paper is organised as follows. In Section 2, we briefly review the Zn Belavin model with
integrable open boundary conditions, which also serves as an introduction to our notion and
basic ingredients. In Section 3, we introduce the intertwiner vectors and the corresponding facevertex correspondence relations which will play key roles in transforming the model in vertex
1 Solutions with arbitrary boundary parameters were recently proposed by functional Bethe Ansatz [17] and q-Onsager
algebra [18]. However, it seems highly non-trivial to rederive these results in the framework of algebraic Bethe Ansatz.
593
picture to the one in the face picture. In Section 4, after finding the n reference states, we use
the algebraic Bethe Ansatz to obtain the corresponding n sets of eigenvalues and the associated
n sets of Bethe Ansatz equations of the transfer matrix of the model. In Section 5, we take
the quasi-classic limit to extract the spectrum of the associated Gaudin operators. Section 6 is
for conclusion. Appendix A provides the definitions of some elementary functions appeared in
Sections 4 and 5.
2. Z n Belavin model with open boundaries
Let us fix a positive integer n 2, a complex number such that Im( ) > 0 and a generic
complex number w. Introduce the following elliptic functions

a
(u, ) =
exp 1 (m + a)2 + 2(m + a)(u + b) ,
b
m=
1 j
(j ) (u) = 2 1 n (u, n ),
(u) = 21 (u, ),
(u) =
(2.1)
(2.2)

ln (u) .
u
(2.3)
Among them the -function2 satisfies the following identity:

(u + x) (u x) (v + y) (v y) (u + y) (u y) (v + x) (v x)
= (u + v) (u v) (x + y) (x y).
Let g, h be n n matrices with the elements
hij = i+1j ,
gij = i ij ,
with = e
2 1
n
, i, j Zn .
For any = (1 , 2 ), 1 , 2 Zn , one introduces an n n matrix I by

I = I(1 ,2 ) = g 2 h1 ,
and an elliptic function (u) by
1 1
+
(u) = 21 n2 (u, ),
2 + n
and
(0,0) (u) = (u).
The Zn Belavin R-matrix is [4,28]

R B (u) =
(w) (u + wn )
I I1 .
(u + w)
n ( wn )
2
(2.4)
Zn
The R-matrix satisfies the QYBE (1.1) and the properties [28],
B
B
Unitarity: R12
(u)R21
(u) = id,
(2.5)
2 Our -function is the -function (u) [29]. It has the following relation with the Weierstrassian -function denoted
1

2
nq 2n
1 . Consequently, our -function (2.3) is
by w (u): w (u) e1 u (u), 1 = 2 ( 16 4
n=1 1q 2n ) and q = e
different from the Weierstrassian -function by an additional term 21 u.
594
B t2
t2
(u) (u + nw)
R 21 (u nw) R B 12
(u) =
id, (2.6)
(u + w) (u + nw w)

B
(u)w0 = id.
Quasi-classical property: R12
(2.7)
Crossing-unitarity:
B (u) = P R B (u)P with P being the usual permutation operator and t denotes the
Here R21
12 12
12
12
i
transposition in the ith space. Here and below we adopt the standard notation: for any matrix
A End(Cn ), Aj is an operator embedded in the tensor space Cn Cn , which acts as A
on the j th space and as an identity on the other factor spaces; Rij (u) is an embedding operator
of R-matrix in the tensor space, which acts as an identity on the factor spaces except for the ith
and j th ones. The quasi-classical properties (2.7) of the R-matrix enables one to introduce the
associated classical Zn r-matrix r(u) [30]

R B (u) = id + wr(u) + O w 2 , when w 0,

(0) (u)
1n
1

(u) +
I I , (0) =
(u) .
r(u) =
(2.8)
n
n (u) (0)
u
Z2n (0,0)
u=0
In the above equation, the elliptic -function is defined in (2.3).

One introduces the row-to-row monodromy matrix T (u) [5], which is an n n matrix with
elements being operators acting on (Cn )N
B
B
B
T (u) = R01
(u + z1 )R02
(u + z2 ) R0N
(u + zN ).
(2.9)
Here {zi | i = 1, . . . , N } are arbitrary free complex parameters which are usually called inhomogeneous parameters. With the help of the QYBE (1.1), one can show that T (u) satisfies the
so-called RLL relation
B
B
R12
(u v)T1 (u)T2 (v) = T2 (v)T1 (u)R12
(u v).
(2.10)
An integrable open chain can be constructed as follows [6]. Let us introduce a pair of
K-matrices K (u) and K + (u). The former satisfies the RE
B
B
R12
(u1 u2 )K1 (u1 )R21
(u1 + u2 )K2 (u2 )
B
B
= K2 (u2 )R12
(u1 + u2 )K1 (u1 )R21
(u1 u2 ),
(2.11)
and the latter satisfies the dual RE

B
B
R12
(u2 u1 )K1+ (u1 )R21
(u1 u2 nw)K2+ (u2 )
B
B
= K2+ (u2 )R12
(u1 u2 nw)K1+ (u1 )R21
(u2 u1 ).
(2.12)
For the models with open boundaries, instead of the standard row-to-row monodromy matrix
T (u) (2.9), one needs the double-row monodromy matrix T(u),
T(u) = T (u)K (u)T 1 (u).
(2.13)
Using (2.10) and (2.11), one can prove that T(u) satisfies
B
B
B
B
R12
(u1 u2 )T1 (u1 )R21
(u1 + u2 )T2 (u2 ) = T2 (u2 )R12
(u1 + u2 )T1 (u1 )R21
(u1 u2 ).
(2.14)
Then the double-row transfer matrix of the inhomogeneous Zn Belavin model with open boundary is given by

(u) = tr K + (u)T(u) .
(2.15)
595
The commutativity of the transfer matrices

(u), (v) = 0,
(2.16)
follows as a consequence of (1.1)(2.6) and (2.11)(2.12). This ensures the integrability of the
inhomogeneous Zn Belavin model with open boundary.
In this paper, we consider a non-diagonal K-matrix K (u) which is a solution to the RE
(2.11) associated with the Zn Belavin R-matrix [31]
K (u)st =
n

(i + u) (s)
(t)
(u) ,w (u).
(i + + u) ,w
(2.17)
i=1
The corresponding dual K-matrix K + (u) which is a solution to the dual RE (2.12) has been
obtained in [32]. With a particular choice of the free boundary parameters with respect to K (u),
we introduce the corresponding dual K-matrix K + (u),

n

((i k ) w) (i + + u + nw
+
s
2 )
K (u)t =
nw
(i k )
(i + u 2 )
i=1 k
=i
(s)
(t)
,w (u) ,w (u).
(2.18)
are intertwiners which will be specified in Section 4. We consider

In (2.17) and (2.18), , ,
the generic {i } such that i
= j (modulo Z + Z) for i
= j . This condition
is necessary for the

non-singularity of K (u). It is convenient to introduce a vector = ni=1 i i associated with
the boundary parameters {i }, where {i , i = 1, . . . , n} is the orthonormal basis of the vector
space Cn such that i , j = ij . Moreover, in the following we always assume that the suffix
index of the parameter i takes value in Zn cyclic group, namely,
in = i ,
i = 1, . . . , n.
(2.19)
3. A(1)
n1 SOS R-matrix and face-vertex correspondence
The An1 simple roots {i } can be expressed in terms of the orthonormal basis {i } as:
i = i i+1 ,
i = 1, . . . , n 1,
and the fundamental weights {i | i = 1, . . . , n 1} satisfying i , j = ij are given by

i =
i

k=1
i
k
k .
n
n
k=1
Set
1
k , i = 1, . . . , n,
n
n
= i ,
=
k=1
n1
then
n
= 0.
(3.1)
i=1
Z+
(the set of non-negative integers), there exFor each dominant weight = i=1 ai i , ai
ists an irreducible highest weight finite-dimensional representation V of An1 with the highest
vector |. For example the fundamental vector representation is V1 .
Let h be the Cartan subalgebra of An1 and h be its dual. A finite-dimensional diagonalizable h-module is a complex finite-dimensional vector space W with a weight decomposition
596

W = h W [], so that h acts on W [] by xv = (x)v, for any x h, v W []. For example, the fundamental vector representation V1 = Cn , the non-zero weight spaces W [ ] = Ci ,
i = 1, . . . , n.
For a generic m Cn , define
mi = m, i ,
mij = mi mj = m, i j ,
Let R(u, m) End(Cn Cn ) be the R-matrix of the

R(u, m) =
n
Riiii (u, m)Eii Eii +

i, j = 1, . . . , n.
(1)
An1
SOS model [33],

ij
ji
Rij (u, m)Eii Ejj + Rij (u, m)Ej i Eij ,
i
=j
i=1
(3.2)
(3.3)
where Eij is the matrix with elements (Eij )lk = j k il . The coefficient functions are
Riiii (u, m) = 1,
ji
Rij (u, m) =
ij
Rij (u, m) =
(w) (u + mij )
,
(u + w) (mij )
(u) (mij w)
,
(u + w) (mij )
i
= j,
i
= j,
(3.4)
(3.5)
and mij are defined in (3.2). The R-matrix satisfies the dynamical (modified) QYBE

R12 u1 u2 , m wh(3) R13 (u1 u3 , m)R23 u2 u3 , m wh(1)

= R23 (u2 u3 , m)R13 u1 u3 , m wh(2) R12 (u1 u2 , m),
(3.6)
and the quasi-classical property

R(u, m)w0 = id.
(3.7)
We adopt the notation: R12 (u, m wh(3) ) acts on a tensor v1 v2 v3 as R(u, m w) id

if v3 W []. Moreover, the R-matrix satisfies the unitarity and the modified crossing-unitarity
relation.
Let us introduce n intertwiner vectors which are n-component column vectors m,mwj (u)
labelled by j (j = 1, . . . , n). The kth element of m,mwj (u) is given by
(k)
m,mwj (u) = (k) (u + nmj ).
(3.8)
We remark that the n intertwiner vectors m,mwj (u) are linearly independent for a generic
m Cn .
Using the intertwining vector, one derives the following face-vertex correspondence relation [33]
B
R12
(u1 u2 )m,mw (u1 ) mw ,mw( +j) (u2 )

R(u1 u2 , m)kl
=
k)
(u1 ) m,mwl(u2 ).
ij mwl,mw(
l+
(3.9)
k,l
Then the QYBE (1.1) of the Zn Belavins R-matrix R B (u) is equivalent to the dynamical Yang
(1)
Baxter equation (3.6) of the An1 SOS R-matrix R(u, m). For a generic m, we may introduce
satisfying the conditions,
other types of intertwiners ,
n

k=1
(k)
(k)
m,mw (u)m,mw (u) = ,
(3.10)
n
(k)
(k)
(u)m+w ,m (u) = ,
m+w,m
597
(3.11)
k=1
from which one can derive the relations,

n

=1
n
(j )
(i)
m,mw (u)m,mw (u) = ij ,
(3.12)
(j )
(i)
(u)m+w,m
(u) = ij .
m+w,m
(3.13)
=1
With the help of (3.10)(3.13), we obtain the following relations from the face-vertex correspondence relation (3.9):
B

m+wk,m
(u1 ) id R12 (u1 u2 ) id m+w j,m (u2 )
R(u1 u2 , m)kl
=
(3.14)
l),m+w
ij m+w( +j),m+wj (u1 ) m+w(k+

k (u2 ),

i,l
m+wk,m
(u1 ) m+w(k+
l),m+w
k (u2 ) R12 (u1 u2 )

=
R(u1 u2 , m)kl
ij m+w( +j),m+wj (u1 ) m+wj,m (u2 ),
(3.15)
i,j
B

(u1 u2 ) m,mw (u1 ) id
id m,mwl(u2 ) R12
R(u1 u2 , m)kl
=
l)
(u1 ) mw ,mw( +j) (u2 ),
ij mwl,mw(
k+

(3.16)
k,j
mwl,mw(
l)
(u1 ) m,mwl(u2 ) R12 (u1 u2 )
k+
=
R(u1 u2 , m)kl
ij m,mw (u1 ) mw ,mw( +j) (u2 ).
(3.17)
i,j
The face-vertex correspondence relations (3.9) and (3.14)(3.17) will play an important role in
translating formulas in the vertex form into those in the face form.
Corresponding to the vertex type K-matrices (2.17) and (2.18), one introduces the following
face type K-matrices K and K [32]
(s)
j
(t)
w( j),w (u)K(u)st ,w (u),
K(|u)i =
(3.18)
s,t
K(|u)
i =

s,t
(s)
st (t)
(u).
,wj (u)K(u)
w(j ),wj
(3.19)
Through straightforward calculations, we find the face type K-matrices simultaneously have diagonal forms3
j
K(|u)i = i k(|u; )i ,
j
j
K(|u)
i = i k(|u)i ,
(3.20)
3 As will be seen below (see (4.27)), the spectral parameter u and the boundary parameter of the reduced double-row
monodromy matrices constructed from K(|u) will be shifted in each step of the nested Bethe Ansatz procedure [21].
Therefore, it is convenient to specify the dependence on the boundary parameter of K(|u) in addition to the spectral
parameter u.
598
where functions k(|u; )i , k(|u)

i are given by
(i + u)
k(|u; )i =
,
(i + + u)
n

(ik w) (i + + u +
k(|u)i =
(ik )
(i + u
k
=i, k=1
(3.21)
nw
2 )
nw .
2 )
(3.22)
Moreover, one can check that the matrices K(|u) and K(|u)
satisfy the SOS type reflection
equation and its dual, respectively [32]. Although the K-matrices K (u) given by (2.17) and
(2.18) are generally non-diagonal (in the vertex form), after the face-vertex transformations
(3.18) and (3.19), the face type counterparts K(|u) and K(|u)
become simultaneously diagonal. This fact enables the authors to apply the generalized algebraic Bethe Ansatz method
developed in [21] for SOS type integrable models to diagonalize the transfer matrix (u) (2.15).
4. Algebraic Bethe Ansatz
By means of (3.12), (3.13), (3.19) and (3.20), the transfer matrix (u) (2.15) can be recast
into the following face type form:

(u) = tr K + (u)T(u)

tr K + (u)w(
=
),w (u)w(
),w (u)
,

T(u),w (u) ,w (u)
,w (u)K + (u)w(
=
),w (u)w(
),w (u)
,
T(u),w (u)

k(|u)
K(|u)
=
T (|u) .
T (|u) =
,
(4.1)
Here we have introduced the face-type double-row monodromy matrix T (|u),

T (|u) = w(
),w (u)T(u),w (u)
(j )
j (i)
(u)T(u)i ,w
(u).
w(
),w
(4.2)
i,j
This face-type double-row monodromy matrix can be expressed in terms of the face type Rmatrix R(u, ) (3.3) and the K-matrix K(|u) (3.18) [21]. Moreover from (2.14), (3.9) and (3.13)
one may derive the following exchange relations among T (m|u) :

i

i j
R(u1 u2 , m)i01 j01 T m + w(j1 + 2 )|u1 i1
2
i1 ,i2 j1 ,j2
j

j i
R(u1 + u2 , m)j12 i23 T m + w(j3 + 3 )|u2 j2
3

j
i j
=
T m + w(j1 + 0 )|u2 j0 R(u1 + u2 , m)i01 j12
1
i1 ,i2 j1 ,j2

i
j i
T m + w(j2 + 2 )|u1 i1 R(u1 u2 , m)j23 i23 .
2
(4.3)
599
Following [21] and motivated by our recent work [20] for the open XXZ chain, we introduce
n sets of operators {A(s) , B (s) , C (s) , D(s) }, labelled by s = 1, . . . , n, as follows:
A(s) (m|u) = T (m|u)ss ,
Bi(s) (m|u) =
T (m|u)si
,
(mis )
Ci(s) (m|u) =
T (m|u)is
,
(msi )
i
= s,
(4.4)

(mj s ij w)
j
j
j
js
D(s) i (m|u) =
T (m|u)i i R(2u, m + w s )s j A(s) (m|u) ,
(mis )
i, j
= s.
(4.5)
Some remarks are in order. Among the n sets of operators, the first set (corresponding to s = 1)
is the very one which was used in [21,23] to construct the algebraic Bethe Ansatz. Such algebraic
Bethe Ansatz based on the first set of operators only gives rise to the first set of eigenvalues and
associated Bethe Ansatz equations. In order to find the complete sets of eigenvalues, we find that
the whole n sets of operators {A(s) , B (s) , C (s) , D(s) | s = 1, . . . , n} are needed.
After tedious calculations analogous to those in [21], we have found the commutation relations
among A(s) (m|u), D(s) (m|u) and B (s) (m|u) from (4.3). Here we give those which are relevant
for our purpose

A(s) (m|u)Bi(s) m + w( s )|v
=

(u + v) (u v w) (s)
Bi m + w( s )|v A(s) m + w( s )|u
(u + v + w) (u v)
(u v msi + w)
(w) (2v)
(u v) (2v + w)
(msi w)
(s)

(s)
Bi m + w( s )|u A m + w( s )|v
(u + v + ms + 2w)
(w)
(u + v + w)
(ms + w)
B(s)
=s

m + w( s )|u D(s) i m + w( s )|v ,
i
= s,
(4.6)

k
D(s) i (m|u)Bj(s) m + w(j s )|v
=
(u v + w) (u + v + 2w)
(u v) (u + v + w)

k

R(u + v + w, m w )2 21 R(u v, m + w j)j 1 i 1

1 ,2 ,1 ,2
=s
(s)
B2 m + w(k

2
+ 2 s )|v D(s) 1 m + w(j s )|u

(u v + ms w)
k
R(2u + w, m w ) i
(ms w)
,
=s

(s)
B m + w(k + s )|u D(s) j m + w(j s )|v
(w) (2u + 2w)

(u v) (2u + w)
600
(w) (2v) (2u + 2w)

(u + v + w) (2v + w) (2u + w)

(u + v + msj )
R(2u + w, m w )kj i
(msj w)
+
=s
B(s)

(s)

m + w(k + s )|u A m + w(j s )|v ,
i, j, k
= s,

Bi(s) m + w( s )|u Bj(s) m + w( + j 2s )|v
=
(4.7)
R(u v, m 2w s )j i B(s) m + w( s )|v B(s) m + w( + 2s )|u ,
,
=s
i, j
= s.
(4.8)
For the special case of s = 1, the above commutation relations (4.6)(4.8) recover those in
[21,23].
In order to apply the algebraic Bethe Ansatz method, in addition to the relevant commutation
relations (4.6)(4.8), one needs to construct a reference state associated with each s, which is
i
(s)
the common eigenstate of the operators A(s) , D(s) i and is annihilated by the operators Ci . In
contrast to the trigonometric and rational cases with diagonal K (u) [6], the usual highestweight state

1
1
0 0,
..
..
.
.
is no longer the pseudo-vacuum state. However, after the face-vertex transformations (3.18)
and (3.19), the face type K-matrices K(|u) and K(|u)

simultaneously become diagonal. This
suggests that one can translate the Zn Belavin model with non-diagonal K-matrices into the cor
responding SOS model with diagonal K-matrices K(|u) and K(|u)
given by (3.18)(3.19).
Then one can construct the pseudo-vacuum in the face language and use the algebraic Bethe
Ansatz method to diagonalize the transfer matrix.
One of the reference states corresponding to the case of s = 1 (or the first one of the n reference states (4.9) below) was found in [21] and yields the first set of eigenvalues of the transfer
matrix. Here, we give the complete n reference states {| (s) () | s = 1, . . . , n}. For each s, we
propose4
(s)
() = (N1)ws ,N ws (z1 ) (N2)ws ,(N1)ws (z2 )
,ws (zN ).
(4.9)
These states (s = 1, . . . , n) depend on the boundary parameters {i } and the inhomogeneous

parameters {zj }, but not on the boundary parameters and . We find that the states given
by (4.9) are exactly the reference states in the following sense,

A(s) ( Nw s |u) (s) () = k(|u; )s (s) () ,
(4.10)
4 Such states played an important role in constructing extra centers of the elliptic algebra at roots of unity [34].
601

i
D(s) j ( Nw s |u) (s) ()

N
(s)
(u + zk ) (u zk )
w
w

i (s)
() ,
= j f (u)k u + ;
2
2 j
(u + zk + w) (u zk + w)
k=1
i, j
= s

(s)
Ci ( Nw s |u) (s) () = 0,

(s)
Bi ( N w s |u) (s) ()
= 0,
(4.11)
i
= s,
(4.12)
i
= s.
(4.13)
Here f (s) (u) is given by

f (s) (u) =
(2u) (s + u + w + )
.
(2u + w) (s + u + )
(4.14)
In order to apply the algebraic Bethe Ansatz method to diagonalize the transfer matrix, we
need to assume N = n l with l being a positive integer [21]. For convenience, let us introduce
a set of integers:
Ni = (n i) l,
i = 0, 1, . . . , n 1,
(4.15)
(i)
and n(n1)
2 l complex parameters {vk | k = 1, 2, . . . , Ni+1 , i = 0, 1, . . . , n 2}. As in the usual
nested Bethe Ansatz method, the parameters {vk(i) } will be used to specify the eigenvectors of
the corresponding reduced transfer matrices. They will be constrained later by Bethe Ansatz
equations. For convenience, we adopt the following convention:
(0)
v k = vk ,
k = 1, 2, . . . , N1 .
(4.16)
We will seek the common eigenvectors (i.e. the so-called Bethe states) of the transfer matrix by
acting the creation operators Bi(s) on the reference state | (s) ()

(s)

(s)
F i1 ,i2 ,...,iN1 Bi1 + w(1 s )|v1 Bi2 + w(1 + 2 2s )|v2
|v1 , . . . , vN1 (s) =
i1 ,...,iN1
=s
(s)
BiN
1

+w
N1

k wN1 s vN1 (s) () .
(4.17)
k=1
The indices in the above equation should satisfy the following condition: the number of ik = j ,
denoted by #(j ), is l, i.e.
#(j ) = l,
j
= s.
(4.18)
Then (3.1) and the above restriction (4.18) imply

+w
N1

k=1
k wN1 s = + wl
n
j wN1 s = w(l + N1 )s = wN s ,
(4.19)
j
=s
which is crucial for the diagonalization of the transfer matrix in the remaining part of the paper.
With the help of (4.1), (4.4) and (4.5) we rewrite the transfer matrix (2.15) in terms of the
i
operators A(s) and D(s) i ,
602
(u) =
n
k(|u)
T (|u)
=1
(s)
= k(|u)
s A (|u) +
k(|u)
i T (|u)i
i
=s
(s)
= k(|u)
s A (|u) +
is (s)
k(|u)
i R(2u, + w s )si A (|u)
i
=s

i
is (s)
k(|u)
i T (|u)i R(2u, + w s )si A (|u)
i
=s
n
is (s)
k(|u)
i R(2u, + w s )si A (|u)
i=1

w (is w)

(1)
(s)
k
T (|u)ii R(2u, + w s )is
u +
+
si A (|u)
2 i (is )
i
=s

w
i

(1)
(s)
(1)
= s (u)A (|u) +
(4.20)
D(s) i (|u).
u +
k
2 i
i
=s
(1)
Here we have used (4.5) and introduced the function s (u),

s(1) (u) =
n
is
k(|u)
i R(2u, + w s )si ,
(4.21)
i=1
and the reduced K-matrix K (1) (|u) with elements given by

j
j
K (1) (|u)i = i k (1) (|u)i , i, j
= s,
n

(ik w) (i + + u +
(1)
k (|u)i =
(ik )
(i + u
k
=i,s
(4.22)
(n1)w
)
2
,
(n1)w
)
2
i
= s.
(4.23)
Using the technique developed in [21], after tedious calculations, we find that with the coefficients F i1 ,i2 ,...,iN1 in (4.17) properly chosen, the Bethe state |v1 , . . . , vN1 (s) becomes the
eigenstate of the transfer matrix (2.15),

(u)|v1 , . . . , vN1 (s) = s u; , {vk } |v1 , . . . , vN1 (s) ,
(4.24)
(i)
provided that the parameters {vk | k = 1, 2, . . . , Ni+1 , i = 0, 1, . . . , n 2} satisfy the following

Bethe Ansatz equations:
s(1) (vj )
N

k=1
(2vj + w)
(2vj + 2w)
N1

k
=j, k=1
(vj + vk ) (vj vk w)
(vj + vk + 2w) (vj vk + w)

(vj + zk ) (vj zk )
w
w (1)
,
+
(1)
;
,
v
v
j
a
(vj + zk + w) (vj zk + w) s
2
2
(4.25)
603
(i)
(i)
(i)
(i)
(i)
Ni+1
(i) (2vj + w)
s(i+1) vj
(i)
(i)
(i)
(i)
(i)
(2vj + 2w) k
=j, k=1 (vj + vk + 2w) (vj vk + w)

(i)
(i)
(i)
(i)
Ni

(vj + zk ) (vj zk )
w (i) w (i+1)
(i)
(i+1)
,
s
vj + ; , v a
=
(i)
(i)
(i)
(i)
2
2
(v + z + w) (v z + w)
k=1
i = 1, . . . , n 2.
(4.26)
(i)
{s (u)
(i)
(i)
{s (u; , {vk })
| i = 1, . . . , n 1} are functions given in Appendix A,

|i=
Here
1, . . . , n 1} are the eigenvalues (given in Appendix A) of the reduced transfer matrices in the
nested Bethe Ansatz process, and the reduced boundary parameters { (i) } and inhomogeneous
(i)
parameters {zk } are given by
(i+1) = (i)
w
,
2
(i+1)
zk
(i)
= vk +
w
,
2
i = 0, . . . , n 2.
(4.27)
(0)
In the above we have adopted the convention: = (0) , zk = zk . The corresponding eigenvalue
s (u; , {vk }) is given by
N1

(u + vk ) (u vk w)
(s + + u + w)
s u; , {vk } = s(1) (u)
(s + + u)
(u + vk + w) (u vk )
k=1
N1
(u vk + w) (u + vk + 2w)
(2u) (s + u + w + )
(2u + w) (s + u + )
(u vk ) (u + vk + w)

(u + zk ) (u zk )
w (1)
w
.
(1)
;
,
v
u
+
i
(u + zk + w) (u zk + w) s
2
2
k=1
(4.28)
k=1
N0

It is easy to check that the first set of eigenvalues 1 (u; , {vk(i) }) and the corresponding Bethe
Ansatz equations are exactly those found in [21]. However, the rest n 1 sets of eigenvalues
{s (u; , {vk }) | s = 2, . . . , n 1} and the associated Bethe Ansatz equations are new ones. For
the special case of n = 2, which corresponds to the open XYZ spin chain, we find that the
two sets of eigenvalues {s (u; , {vk }) | s = 1, 2}, after rescaling of an overall factor due to
different normalizations of the R- and K-matrices, recover those in [35] obtained by directly
solving the T Q relation (or functional Bethe Ansatz). As shown in [35], these two sets of
eigenvalues give the complete spectrum of the transfer matrix of the open XYZ spin chain. As
a consequence, the corresponding two sets of Bethe states {|v1 , . . . , vN1 (s) | s = 1, 2} together
constitute the complete eigenstates of the transfer matrix of the open XYZ model. Therefore,
it is expected that the n sets of eigenvalues {s (u; , {vk }) | s = 1, . . . , n} (4.28) [respectively
Bethe states {|v1 , . . . , vN1 (s) | s = 1, . . . , n} (4.17) and (4.25)(4.26)] together give rise to the
complete spectrum [respectively the complete eigenstates] of the transfer matrix (2.15) of the
open Zn Belavin model.
5. Result for the associated Gaudin model
As will be seen from the definitions of the intertwiners (3.8), (3.10) and (3.11), specialized to
m = , ,w (u) and ,w (u) do not depend on w while ,w (u) does. Consequently,
the K-matrix K (u) does not depend on the crossing parameter w, but K + (u) does. So we use
604
the convention:
K(u) = lim K (u) = K (u).
(5.1)
w0
We further assume that the parameter has the following behavior5 as w 0,

= + w + O w 2 ,
(5.2)
with a constant . It implies that

lim = .
w0
Then the K-matrices satisfy the following relation

lim K + (u)K (u) = lim K + (u) K(u) = id.
w0
w0
(5.3)
Let us introduce the elliptic Gaudin operators {Hj | j = 1, 2, . . . , N } associated with the inhomogeneous Zn Belavin model with open boundaries specified by the generic K-matrices (2.17)
and (2.18):
N

N

1
Hj = j (zj ) +
(5.4)
rkj (zj zk ) + Kj (zj )
rj k (zj + zk ) Kj (zj ),
k
=j
k
=j
B (2u)P }.
where j (u) = w
{K j (u)}|w=0 Kj (u), j = 1, . . . , N , with K j (u) = tr0 {K0+ (u)R0j
0j
Here {zj } are the inhomogeneous parameters of the inhomogeneous Zn Belavin model and r(u)
is given by (2.8). For a generic choice of the boundary parameters {1 , . . . , n , }, j (u) is
a non-diagonal matrix.
Following [36,37], the elliptic Gaudin operators (5.4) are obtained by expanding the doublerow transfer matrix (2.15) at the point u = zj around w = 0:

(zj ) = (zj )w=0 + wHj + O w 2 , j = 1, . . . , N,
(5.5)

.
(zj )
Hj =
(5.6)
w
w=0
The relations (2.7) and (5.3) imply that the first term (zj )|w=0 in the expansion (5.5) is equal to
an identity, namely,

(zj )w=0 = id.
(5.7)
Then the commutativity of the transfer matrices { (zj )} (2.16) for a generic w implies
[Hj , Hk ] = 0,
i, j = 1, . . . , N.
(5.8)
Thus the elliptic Gaudin system defined by (5.4) is integrable. Moreover, the relation (5.6) between {Hj } and { (zj )} and the fact that the first term on the r.h.s. of (5.5) is identity operator
enable us to extract the eigenstates of the elliptic Gaudin operators and the corresponding eigenvalues from the results obtained in the previous section.
5 In [12,20,23], a special case of = was studied. The generalization to the case with nonvanishing is straightforward.
605
Using (4.27), (A.3), (A.4) and (A.6)(A.8), the Bethe Ansatz equations (4.25) and (4.26)
become, respectively,
(i)
(i)
(i)
(i)
(i)
Ni+1
(2vj + w)
s(i+1) vj(i)
(i)
(i)
(i)
(i)
(i)
(2vj + 2w) k
=j, k=1 (vj + vk + 2w) (vj vk + w)

(i)
vj
= s(i+2)

(i+1) + v (i) + 3 w)
w (i+s+1 +
j
2
+
(i)
(i+1)
2 (i+s+1 +
+ vj + 12 w)
(i)
(i1)
Ni

(vj + vk
+
(i)
k=1
(i1)
(vj + vk
(i)
(i1)
w
w2 )
2 ) (vj vk
(i)
(i1)
3w
+ w2 )
2 ) (vj vk
Ni+2
(vj(i) + vk(i+1) + w2 ) (vj(i) vk(i+1) w2 )
,
(i)
(i+1)
(i)
(i+1)
+ 3w
+ w2 )
k=1 (vj + vk
2 ) (vj vk
i = 0, . . . , n 3,
(5.9)
(n2)
n1
+ w) N

(2vj
(n1) (n2)
vj
s
(2vj(n2) + 2w) k
=j, k=1
(s1 + + vj
(n2)
(vj(n2) + vk(n2) ) (vj(n2) vk(n2) w)

(vj(n2) + vk(n2) + 2w) (vj(n2) vk(n2) + w)
(n2)
+ w) (s1 + (n1) vj
(n2)
(n2)
(s1 + vj
w) (s1 + (n1) + vj
+
(vj(n2) + vk(n3) +
Nn2
k=1
(vj(n2) + vk(n3) +
w
2)
w
2)
(n2)
(n3)
w
vk
w2 )
2 ) (vj
.
(n2)
3w
vk(n3) + w2 )
2 ) (vj
(5.10)
(1)
Here we have used the convention: vk = zk , k = 1, . . . , N . The quasi-classical property (3.7)

of R(u, m), (A.1) and (A.2) lead to the following relations
s(i+1) (u, , 0) = 1,
(i+1)
(u, , 0) = 0,
u s
i = 0, . . . , n 2.
(5.11)
(i+1)
(u)} associated with

Noticing the restriction (5.2), one may introduce some functions {s
(i+1)
(u, , w)},
{s

(i+1)
(i+1)
(i+1)

s
(u) =
(u, , w)
+ s
(u, , 0) , i = 0, . . . , n 2.
w s
w=0
=
(5.12)
(s)
Using (5.5), we can extract n sets of eigenvalues {hj | s = 1, . . . , n} (respectively the corresponding Bethe Ansatz equations) of the Gaudin operators Hj (5.4) from the expansion around
w = 0 for the first order of w of the eigenvalues (4.28) of the transfer matrix (u = zj ) (respectively the Bethe Ansatz equations (5.9) and (5.10)). Finally, the eigenvalues of the Zn elliptic
Gaudin operators are
(s)
hj = s(1) (zj ) + (s + + zj )
N1

(zj + xk ) + (zj xk ) ,
k=1
(5.13)
606
(i)
where -function is defined in (2.3). The n(n1)l

parameters {xk | k = 1, 2, . . . , Ni+1 , i =
2
(0)
0, 1, . . . , n 2} (including xk as xk = xk , k = 1, . . . , N1 ) are determined by the following
Bethe Ansatz equations
s(i+1)
(i)
(i)
xj 2xj 2
(i)
(i)
(i)
(i)
xj + xk + xj xk
Ni+1
k
=j, k=1
(i)
= s(i+2) xj
Ni
(i)
(i)
(i1)
(i1)
xj + x k
+ xj xk
k=1
i+2

(i)
(i)

(i)
(i+1)
(i+1)
xj + x k
+ xj xk
,
+ i+s+1 + + xj
k=1
i = 0, . . . , n 3,
(5.14)
(n2)
(n2)
(n2)
(n2)
(n2)
(n2)
s(n1) xj
2xj
2
xj
+ xj
+ xk
xk
Nn1
k
=j, k=1

n
2n
(n2)
(n2)
= +
+
s1 + + xj
s1 + xj
2
2
(n2)
(n2)
(n3)
(n3)
xj
+ xj
.
+ xk
xk
Nn2
(5.15)
k=1
(1)
= zk , k = 1, . . . , N , in (5.14). Then the n sets of eigenHere we have used the convention: xk

(s)
values {hj | s = 1, . . . , n} given by (5.13)(5.15) (cf. [23]) together constitute the complete
spectrum of the Gaudin operators Hj (5.4).
6. Conclusions
We have discovered the multiple reference state structure of the Zn Belavin model with boundaries specified by the non-diagonal K-matrices K (u), (2.17) and (2.18). It is found that there
exist n reference states {| (s) () | s = 1, . . . , n} (4.9), which lead to n sets of Bethe states
|v1 , . . . , vN1 (s) (4.17). These Bethe states give rise to n sets of Bethe Ansatz equations (4.25)
(4.26) and eigenvalues (4.28), labelled by s = 1, . . . , n. The fist set of them, which corresponds
to the s = 1 case, gives the results found in [21]. It is expected that these n sets of eigenvalues
{s (u; , {vk }) | s = 1, . . . , n} (4.28) together give rise to the complete spectrum of the transfer matrix (u) (2.15) for the Zn Belavin model with generic boundaries. In the quasi-classical
(s)
limit (i.e. w 0), the resulting n sets of eigenvalues {hj | s = 1, . . . , n} given by (5.13)(5.15)
together constitute the complete spectrum of the Gaudin operators Hj (5.4).
Taking the scaling limit [38] of our general results, for the special n = 2 case, we recover the
results obtained in [20] for the open XXZ spin chain. It is believed that such structure of multiple
reference states also exists for the open A(1)
n1 trigonometric vertex model studied in [24].
Acknowledgements
The financial support from Australian Research Council is gratefully acknowledged.
607
Appendix A. Definitions of the -, -, -functions

(i)
(i)
(i)
In this appendix, we give the definitions of the functions s (u), s (u) and (i) (u; , {vk }),
which appeared in the expressions of the eigenvalues and the Bethe Ansatz equations (4.25)
(4.28).
In order to carry out the nested Bethe Ansatz for the Zn Belavin model with the generic open
boundary conditions, one needs to introduce a set of reduced K-matrices K (r) (|u) which include
the original one K(|u)

= K (0) (|u):
j
j
K (r) (|u)i = i k (r) (|u)i , i, j = r + 1, . . . , n, r = 0, . . . , n 1,

n

)
(i + + u + (nr)w
w)
ik
2
k (r) (|u)i =
.
(nr)w
(ik )
(i + u
)
k
=i, k=r+1
Moreover we introduce a set of functions

K-matrices K (r) (|u)
(r) (u) =
n
{ (r) (u)
(r1) (|u)i ,
R(2u, + w r )ir
ri k
| r = 1, . . . , n 1} related to the reduced
r = 1, . . . , n 1,
(A.1)
i=r
and an associated set of functions { (i) (u, , w) | i = 1, . . . , n 1}

(i+1) (u, , w) (i+1) (u) = (i+1) (u)
(i+1 + u 2i w)
(i+1 + + u + w 2i w)
i = 0, . . . , n 2.
(A.2)
In the process of carrying out the nested Bethe Ansatz [21], one needs to introduce a set of
(i)
functions {(i) (u; , {vk }) | i = 0, . . . , n 1} which correspond to the eigenvalues of the re(i)
duced transfer matrices. The functions {(i) (u; , {vk })} are given by the following recurrence
relations

(i) u; (i) , vk(i)
= (i+1) (u)
Ni+1
(i)
(i)
(i+1 + (i) + u + w) (u + vk ) (u vk w)
(i)
(i)
(i+1 + (i) + u)
k=1 (u + vk + w) (u vk )
Ni+1
(i)
(i)
(2u) (i+1 + u + w + (i) ) (u vk + w) (u + vk + 2w)
(i)
(i)
(2u + w) (i+1 + u + (i) )
(u vk ) (u + vk + w)
k=1

Ni
(i)
(i)

(u + zk ) (u zk )
w (i) w (i+1)
(i+1)
,
u + ; , vj
(i)
(i)
2
2
k=1 (u + zk + w) (u zk + w)
i = 1, . . . , n 2,

(n + + u +
(n1) u; (n1) =
(n + u
(A.3)
w
2 ) (n
w
2 ) (n
+ (n1) u)
+ (n1) + u)
(A.4)
The reduced boundary parameters { (i) } and inhomogeneous parameters {zk(i) } are given by
(i)
(4.27). It is remarked that all the functions { (i) (u)}, { (i) (u)} and {(i) (u; , {vk })} are indeed the functions of the boundary parameters {i }.
608
Since that the suffix index of the boundary parameters {i } takes value in Zn , one can introduce a Zn cyclic operator P (i.e. P n = id), which acts on the space of functions of {1 , . . . , n }.
On any function f (1 , . . . , n ) the action of the operator P is given by

P f (1 , . . . , n1 , n ) = f (1+1 , . . . , n , n+1 ) = f (2 , . . . , n , 1 ).
(A.5)
Then we introduce the following -, -, -functions:

s(i) (u) = P s1 (i) (u) , s = 1, . . . , n, i = 1, . . . , n 1,

s(i) (u) = P s1 (i) (u) , s = 1, . . . , n, i = 1, . . . , n 1,

(i)

(i)
(i)
= P s1 (i) u; , vk , s = 1, . . . , n,
s u; , vk
(A.6)
(A.7)
i = 1, . . . , n 1.
(A.8)
(1)
Direct calculation shows that the two definitions of s , (4.21) and (A.6), coincide with each
other.
References
[1] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic, London, 1982.
[2] M.A. Kastner, R.J. Birgeneau, G. Shirane, Y. Endoh, Rev. Mod. Phys. 70 (1998) 897.
[3] L. Dolan, C.R. Nappi, E. Witten, JHEP 0310 (2003) 017;
B. Chen, X.J. Wang, Y.S. Wu, Phys. Lett. B 591 (2004) 170;
M. Kruczenski, Phys. Rev. Lett. 93 (2004) 161602.
[4] A. Belavin, Nucl. Phys. B 180 (1981) 189.
[5] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation Function, Cambridge Univ. Press, Cambridge, 1993.
[6] E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
[7] L. Mezincescue, R.I. Nepomechie, J. Phys. A 24 (1991) L17.
[8] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841, hep-th/9306002.
[9] R.I. Nepomechie, J. Stat. Phys. 111 (2003) 1363, hep-th/0211001;
R.I. Nepomechie, J. Phys. A 37 (2004) 433, hep-th/0304092.
[10] J. Cao, H.-Q. Lin, K.-J. Shi, Y. Wang, Nucl. Phys. B 663 (2003) 487.
[11] A. Doikou, Nucl. Phys. B 668 (2003) 447, hep-th/0303205;
A. Doikou, J. Stat. Mech. P09 (2006) 010, hep-th/0603112.
[12] W.-L. Yang, Y.-Z. Zhang, M. Gould, Nucl. Phys. B 698 (2004) 503, hep-th/0411048.
[13] J. de Gier, A. Nichols, P. Pyatov, V. Rittenberg, Nucl. Phys. B 729 (2005) 387, hep-th/0505062;
J. de Gier, F.H.L. Essler, Phys. Rev. Lett. 95 (2005) 240601, cond-mat/0508707;
A. Nichols, J. Stat. Mech. L02 (2006) 004, hep-th/0512273.
[14] C.S. Melo, G.A.P. Ribeiro, M.J. Martins, Nucl. Phys. B 711 (2005) 565, nlin.SI/0411038.
[15] W.-L. Yang, R.I. Nepomechie, Y.-Z. Zhang, Phys. Lett. B 633 (2006) 664, hep-th/0511134.
[16] Z. Bajnok, J. Stat. Mech. P06 (2006) 010, hep-th/0601107.
[17] R. Murgan, R.I. Nepomechie, C. Shi, J. Stat. Mech. P08 (2006) 006, hep-th/0605223;
R. Murgan, R.I. Nepomechie, C. Shi, JHEP 0701 (2007) 038, hep-th/0611078.
[18] P. Baseilhac, K. Koizumi, Exact spectrum of the XXZ open spin chain from q-Onsager algebra representation
theory, hep-th/0703106.
[19] R.I. Nepomechie, F. Ravanini, J. Phys. A 36 (2003) 11391, hep-th/0307095;
R.I. Nepomechie, F. Ravanini, J. Phys. A 37 (2004) 1945, Addendum.
[20] W.-L. Yang, Y.-Z. Zhang, JHEP 0704 (2007) 044, hep-th/0703222.
[21] W.-L. Yang, R. Sasaki, Nucl. Phys. B 679 (2004) 495, hep-th/0308127.
[22] W. Galleas, M.J. Martins, Phys. Lett. A 335 (2005) 167, nlin.SI/0407027.
[23] W.-L. Yang, R. Sasaki, Y.-Z. Zhang, JHEP 0409 (2004) 046, hep-th/0409002.
[24] W.-L. Yang, Y.-Z. Zhang, JHEP 0501 (2005) 021, hep-th/0411190;
W.-L. Yang, Y.-Z. Zhang, JHEP 0412 (2004) 019, hep-th/0411160.
[25] W.-L. Yang, Y.-Z. Zhang, R. Sasaki, Nucl. Phys. B 729 (2005) 594, hep-th/0507148.
[26] W. Galleas, Spectrum of the supersymmetric tJ model with non-diagonal open boundaries, nlin.SI/0703003.
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
609
M. Gaudin, J. Phys. (Paris) 37 (1976) 1087.

M.P. Richey, C.A. Tracy, J. Stat. Phys. 42 (1986) 311.
E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, fourth ed., Cambridge Univ. Press, Cambridge, 2002.
B.Y. Hou, W.-L. Yang, J. Phys. A 32 (1999) 1475, q-alg/9711010.
H. Fan, B.Y. Hou, G.L. Li, K.J. Shi, Phys. Lett. A 250 (1998) 79.
W.-L. Yang, R. Sasaki, J. Math. Phys. 45 (2004) 4301, hep-th/0308118.
M. Jimbo, T. Miwa, M. Okado, Lett. Math. Phys. 14 (1987) 123;
M. Jimbo, T. Miwa, M. Okado, Nucl. Phys. B 300 (1988) 74.
W.-L. Yang, A. Belavin, R. Sasaki, Nucl. Phys. B 710 (2005) 614, hep-th/04004089.
W.-L. Yang, Y.-Z. Zhang, Nucl. Phys. B 744 (2006) 312, hep-th/0512154.
K. Hikami, J. Phys. A 28 (1995) 4997.
E.K. Sklyanin, T. Takebe, Phys. Lett. A 219 (1996) 217.
W.-L. Yang, Y. Zhen, Commun. Theor. Phys. 36 (2001) 131, math-ph/0103028.

Nucl - Phys.B v.789

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.789

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 789 (2008) 144

Glueballs vs. gluinoballs: Fluctuation spectra in

via Cintia, 80126 Napoli, Italy

E-mail address: mberg@aei.mpg.de (M. Berg).

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

where the background scalars, (r),

M. Berg et al. / Nuclear Physics B 789 (2008) 144

with constants ci and ci . Thus, (2.10) implies that

interesting to find more precise conditions for these statements.

M. Berg et al. / Nuclear Physics B 789 (2008) 144

from which one can deduce the orthogonality relation

from which follows that

2.3. Pole structure of holographic 2-point functions

0|O1 (0)||O2 (0)|0

M. Berg et al. / Nuclear Physics B 789 (2008) 144

Oi (k) exact = Yij k 2 dj k 2 + local terms,

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

By symmetry, we find that Yij equals Zij up to normalization,

Therefore, defining also

our final result is

M. Berg et al. / Nuclear Physics B 789 (2008) 144

3.1. Mass spectrum: Determinant method

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

and the superpotential16

M. Berg et al. / Nuclear Physics B 789 (2008) 144

from which follow N a 3 = N 3 b = 0. Moreover, one easily checks that G 3 bc = G a 3c = 0, so that

which is identical to (4.10).

M. Berg et al. / Nuclear Physics B 789 (2008) 144

takes one of the values [11]

The spectra (4.15) and (4.16) can be combined into

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

which leads to the mass values

M. Berg et al. / Nuclear Physics B 789 (2008) 144

The sigma-model metric is

and the superpotential reads

e2x/34p = 2P 2 e0 32/3 h( ) sinh4/3 ,

Moreover, one can show that the warp function A satisfies

M. Berg et al. / Nuclear Physics B 789 (2008) 144

5.2. KS spin-2 spectrum

M. Berg et al. / Nuclear Physics B 789 (2008) 144

A least-square fit yields

term enjoys a certain degree of

5.3. KS 7-scalar system: Setup

M. Berg et al. / Nuclear Physics B 789 (2008) 144

As before, we fix the momentum scale as in (5.15).

22 The coefficients may contain rational functions of , but no exponentials.

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

in Fig. 4. More values are given in

M. Berg et al. / Nuclear Physics B 789 (2008) 144

m2 0.271n2t + 1.150nt + 0.648,

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

M. Berg et al. / Nuclear Physics B 789 (2008) 144

0|O1 (0)||O2 (0)|0

Oi exact = 2i di = (2i d)di .